Colony Personal Browser – Part 1 - Overview

After researching and implementing versioning for Argot (see recent posts), I've spent a lot of time looking for which direction to head in next. The problem I find with Argot is that there are so many different directions it can be taken. There's so much work that can be done, yet due to time/money constraints I've only got an hour or two a day (while travelling to/from work) that progress can be made.

For a while I have been having a lot of trouble finding what to focus on and jumped between a few projects. I updated Colony to work with Argot versioning. I also rebuilt the software that Argot was originally built; a network monitoring solution that built on a virtual network of nodes for processing and filtering log messages at high speed. I contemplated polishing Argot 1.3; fixing up documentation, adding more examples and improving the web site. I've also thought about the BORED protocol (see older posts) and the redesign of the Colony communication protocol. This has lead to thinking about the model that Argot and Colony were based. The original model dates back over 15 years and hasn't changed much since then.

After a lot of heading off on different tangents, I think I'm now heading in the right direction. I'm looking at the full protocol stack from transport to application layer. I'm returning to the original model that Colony was based and re-establishing the fundamental concept of the personal browser. This will provide a more solid foundation from which to build Colony and the Argot Remote Type Negotiation system. It will also take on some of my learning's from investigating REST with the BORED (Binary Object REst Distributed) protocol.

In the numerous posts regarding BORED protocol, one of the most pertinent things I discovered is that designing protocols requires a very clear understanding of the vision for the protocol. The specific aims, tasks and model must be understood clearly from the start. For this reason its worth revisiting the history and thoughts behind Colony and Argot and understanding the model it's based upon.

Colony History

As I said before, the original model dates back over 15 years to the early 1990's from when I had just started computer science at University. Before I started at Uni I already had a fascination with protocols and communication. This interest was born from using Bulletin Board Systems (BBS's) as a teenager. Before starting Uni I had designed protocols and developed software for drawing vector graphics so two people could both draw on each others screens. What annoyed me from that early age about BBS's was the inability to do more than one thing. As I learnt more about communications I started to develop a model for distributed applications.

The model for Colony was based on a few simple constructs; zones, realms and nodes. These constructs still exist in Colony today, however, they don't integrate as cleanly into the design as I would like. I'll explain each of these constructs and their behaviour now.

A zone is a container with simple name/value pairs. A zone could be implemented as a HashMap, overlay a directory structure or proxy another zone from a remote system. It can contain simple values, objects or other zones. A zone provides the basis for naming and containment in the system.

A realm is an extension of a zone. A realm provides the security aspects of a zone or set of zones. The idea of a realm was that it be implemented as a process but include a security model to ensure applications only had a specific access to the underlying system.

A node provided the processing aspect of the design. A node contains a queue for receiving messages and a thread pool for processing the messages. A distributed application would consist of multiple nodes passing messages. The design was suited to network applications that would process a lot of data from different sources.

Applications would be designed with all their data contained in zones, realms and nodes which would allow real-time introspection of the data and how nodes were behaving.

The original design was based on a message processing architecture. Each node would be designed to perform a small and specific task and forward on the same or new message to another node. Applications could be configured by putting different nodes into a zone and wiring them up to perform a specific function. Messages could be passed between realms to create a distributed application. Since this original design a lot of things have changed about what is important in distributed applications.

Colony Architecture Revisited

Before launching into protocol design based on the original model of Colony, it's worth looking at the original design with over 15 years of experience. The original design was aimed at creating a BBS client/server system that allowed multiple application to communicate concurrently. It was expected that each application would be downloaded and installed and use the naming and communication system of Colony. Since that time things have changed considerably.

TCP/IP and the browser have become the key building block for the largest percentage of web communication. Many of the building blocks for distributed systems have been developed and are a lot better understood. Virtual Machines and scripting languages which leads to code on demand are now available. This all leads to a different architecture for Colony which mixes the old and new.

The revised aim and architecture of Colony is to build a “personal browser”. The Colony Personal Browser mixes concepts of Instant Messaging, Code on Demand, Distributed Computing, Security, Identity and Web Browsing to create a secure connected peer-to-peer browser environment. This environment should allow applications to create secure connections between both individuals and Colony servers offering services and/or web style documents. At a high level the components of the “personal browser” include:



Other than Identity all of these components make up a modern browser; so before moving on I'll explain what is different about each of these components.

Documents & Code on Demand – In a paper I wrote a number of years ago, I outlined a method of using Argot as a common data format for virtual machine byte code, document formatting and as a scripting language. All of these formats are compiled to a common byte code which is then targeted to a more specific virtual machine. Creating a common format for all data creates a very flexible development environment where different styles of programming languages can be built on the same virtual machine. It should also allow mixing different programming styles closely together allowing the best method of development to be used for each problem. You can read the evolvable programming languages paper here.

Security & Identity – The web of trust security model has been well documented and around for many years. However, after the initial interest in PGP its use has dwindled and it has not been accepted into the main stream. The aim of the personal browser is to wrap a set of user interfaces and processes around the web-of-trust model to make it easier and quicker for users to understand and use. You can read more about the web of trust at wikipedia.

Communications – The area of communications is obviously the most important aspect of the personal browser. Communicating between users and servers requires a hybrid peer-to-peer system. This style of communication requires a flexible transport layer that allows any peer to act as a client and initiate requests to other peers. This then requires a flexible naming system that allows peers to be found and communicated. The naming and routing aspect of the personal browser is probably the most difficult with the least defined solution. I'll discuss the protocol in more detail a little later.

Virtual Machine – The virtual machine and underlying execution environment must provide a sandbox environment that ensures the security of the underlying system. This is a well understood area and selecting the Google Chrome environment using V8 or Java Virtual Machine as a basis will both achieve the aims here.

User Interface – This is another area with very well understood and well developed code. Taking a browser layout engine that can be removed from an existing browser without being infected with HTML or CSS will provide a good clean user interface as a starting point.


Colony communication building blocks
This next section will examine some of the building blocks required as input into the next version of the Colony protocol.

Sand Boxes
The concept of realms can be compared to the sand boxes used by both Java and Google Chrome. Google Chrome is interesting in that it separates the browser into multiple Processes to improve security. Colony can use this same mechanism to separate each realm and provide limited access to hosted applications. This can be achieved by using a gateway system which provides firewall and routing mechanisms to data being sent and received from applications.



The gateway is able to handle connections to multiple hosts and direct messages between multiple Realms running on the same computer. Between the Gateway and Realms a system pipe or other communication mechanism can be used. For external communications IP based transport will be used.

SCTP – The right transport
One of the big challenges that I've been faced with over the years is building the right type of protocol over TCP. As TCP is a stream based protocol it is faced with a number of issues. A friend recently pointed me towards SCTP as a potentially better transport protocol. After reviewing some of the various introductory web sites on SCTP, I'm convinced. SCTP creates a much better base to build higher level protocols than both UDP or TCP. SCTP provides:

• Multiple Streams on a single connections
• Multi-homing
• Unordered delivery
• Reliable transmission
• etc

There's more information on SCTP available out there. There is also the issue that SCTP is only being introduced in Java 7 which is still in beta. Also, Java 7 when released will not support SCTP on Windows. There is third party implementations of SCTP for windows, however, the protocol is not yet supported by Microsoft. Given I'm developing Colony as a longer term solution I don't believe this will be a problem.

I will publish some prototype code soon which demonstrates TLS on SCTP using Java 7 which I developed on Open Solaris using VMWare.

TLS – Web of Trust

Another big issue in distributed applications is security and identity. To provide security the obvious answer is TLS which is a proven secure protocol. As mentioned earlier, the solution for identity will be the Web-of-trust model.

Ideally the digital certificate data used in the web-of-trust model will be encoded using Argot instead of ASN.1 to ensure consistency across the system. The Web-of-trust model will be developed after the main protocol.

Naming & Location

The URI has proved itself as one of the most flexible solutions to object naming and location. People understand it and find it easy to use. It is not perfect for this application as it binds the name to a specific host and IP address. This may not be a perfect mechanism to use, however, initially it provides the best choice for naming and location. I'll discuss why the URI is not perfect a little later into the design.

Another possible solution to naming and location would be based on a peer-to-peer naming system. This would fit better with the distributed nature of the system.

Protocol Stack Design

The main ingredients of Sand Boxes, SCTP, TLS/Web-of-trust and URI naming are all the ingredients required to design the lower levels of the protocol stack. I'll now examine how these parts would be best glued together in a protocol to meet the design.

My initial thought process for the protocol stack was that it would be obvious that TLS would sit over SCTP to provide secure communications between systems. However, after delving further into the design I realise that using the model where a gateway is the external communications end point it fails to provide true end to end security to the application. To provide end to end security the TLS connection must terminate at the realm and not at the gateway. To support this, a protocol gateway layer (Colony Routing Protocol) is required which acts as an application message router. The protocol stack then looks like:



In effect what is being created is a high level transport protocol that allows messages to be delivered directly to objects with security. This change in model has a number of advantages. It allows the gateway to act as a load balancer or facade onto a group of larger systems. It can also allow the gateway to proxy requests to other gateways. The gateway can also use other transport mechanisms to reach the final Realm destination. On a local computer it can use pipes, while on a remote system it could use SCTP, TCP or other transport. The protocol should also be light weight enough to send through to embedded devices.

As mentioned before, using SCTP encourages a message based system. The gateway router mechanism also suggests that a message based system would also be appropriate for this layer of the protocol stack. A secondary restriction is that TLS requires that each message packet be no longer than 16kb. This puts a restriction on the size of the overall SCTP packet to 16kb plus any header information.

It is expected that the end point location for a message will be any URL. The router will find the realm that contains the target object and direct the message to that realm. For example, a target of crl://some.server/target/realm/object will be received by the realm container at crl://some.server/target/realm.

Another advantage to this model is that while it offers a lot of flexibility, it does so without putting any restriction on the actual protocol used by the Realm or target object. Different objects can use different protocols on the same transport.

The Colony Routing Protocol data packet is likely to contain the following parts:

Preamble – header signifying the colony routing protocol.
Version – version of the protocol. Major & minor version details.
Headers – Additional header data. Does message require response, etc.
Target Location – URI of the target location.
Data – data to be delivered to the realm. Max. size 16kb.
Digital Signature – Optional signature for message.

An interesting element of this protocol design is that it does not dictate request/response semantics as used in REST/HTTP. A client can send a one way message to a target location. It is up to the semantics of the session layer protocol to decide if a response is required. The underlying system also puts no restriction on whether the client or server initiates the request.

To support server initiated requests requires that each client has a host and realm that will receive messages. As clients do not usually have a fully qualified host name it will require an alias is given to the client connection on the server. This is where a traditional HTTP browser paradigms do not provide enough flexibility and a peer based host naming system may be more appropriate.

Another requirement for the routing protocol is to allow the creation of routes between two realms/objects that are expected to last a long period. The advantage of this is that the target location and any digital signatures can be dropped allowing less overhead per packet. This would be especially useful for voice or video communication protocols. It would also allow some of the benefits of SCTP to be exposed to the upper layer protocols. This requires that the communication layer keep session state. Keeping session state is not ideal at this layer of the protocol stack, however, is inevitable to support this type of feature. The state should only be limited to link information state which is no different to the way a NAT router holds state.

The routing layer has many parallels with peer-to-peer naming and routing systems which have been documented previously. This area will need to be researched more thoroughly before locking in a specific solution.

Before finishing the subject of routing and naming it is important not to forget firewalls and NAT. This is the enemy of any peer to peer based system. There are various solutions to this problem which have been documented. A possible solution is to use Colony application servers which act as Internet routers and Proxies for clients behind firewalls. This is another area which will require further investigation.

Session (Discovery) Layer
The session layer will sit on the TLS layer to provide elements of a REST architectural style; specifically the client/server, stateless and uniform interface constraints. The aim of this layer is to provide a discovery, reflection and basic set of mechanisms to communicate with the target objects. The discovery aspect of the protocol layer will allow a client to discover what protocols can be used with the target object. The server could allow multiple presentation and application layer protocols to be provided, allowing a client to choose the most appropriate method to communicate.

The Uniform Interface would include at least the following types of messages:

GET - return the selected or default presentation of the target object. This may be a image or document file. It may also be code on demand.

META – return the associated protocols and presentations available for the target object. This could looking something like:

(protocol:cache,stream data-type:image/jpg)
or
(protocol:dynamic-argot-dictionary application:colony-vm)

DATA – send data to the target object using a selected protocol.

The packet structure for the stream includes the following:

message type – GET, META or DATA. Most likely encoded as a byte.
Protocol type – the protocol type being sent.
Data – data to be passed to the target object.

Depending on the protocol selected there may be multiple session, presentation and application layer options over this transport layer. These should be registered with the target object meta information. The client is able to select the most suitable method to communicate.

Application Level Protocols

The aim of this layered protocol design is to expose as many of the features of the underlying transport to the upper layers of the protocol stack and therefor distributed applications. It is likely that a number of application level protocols would be developed for the most common situations; the most obvious being a stream based solution for serving static files. This would allow both client or server side cache mechanisms to be included. Message queue based protocols are another obvious possible protocol. Smart agent or mobile code as already used by Colony is another possible application layer protocol.

As the design exposes the underlying SCTP packet structure to the application a combination of these protocols are possible. Using SCTP out-of-order and non-guaranteed delivery options also allows video or voice data protocols to be established.

Conclusion
This is very rough draft of the Colony Personal Browser model being developed and how the protocol would be designed to support the model. It builds on concepts from REST, uses SCTP and includes TLS security based on the web of trust model.

The design provides the following features:

Peer to Peer – Application protocols can direct messages to any object using a URL style format that users can easily understand.
User to User – Applications & Documents can be developed which connect users to users in the same way instant messenger applications work.
User to Server – Applications can be developed which connect directly to server applications in the same way the traditional browser operates.
Code & Data on demand – Using the virtual machine for both data and code creates a highly flexible environment to build interactive documents or applications.
Flexible – Application protocols have full access to the flexibility of the SCTP protocol; allowing to choose out of order delivery or non guaranteed delivery. This allows video and voice applications to be built.
Packet Based – Application protocols must adhere to sending a maximum data packet size of 16kb to adhere to TLS. To send larger data packets higher layer protocols must be established.
Short Messages – Messages shorter than 16kb can be sent directly to another object.
Established Links – Long conversations between peers can be created over established links.
Reflection – Objects can publish meta data to describe the available protocols or data presentation formats available.
Simplicity – A user is able to still add a simple text URL into a browser address bar and have a default data representation returned. This ensures the simplicity of web browsers is retained.
Complexity – An application developer can select the best protocol design for the interaction of client and server.
Security – TLS provides Realm to Realm security on all links.
Identity – Where a user chooses, their identity can be made known to the server they are connecting. This is perfect for peer to peer or applications where identity is required.

Obviously this is a rough outline and the details of all the components needs to be fleshed out. There are numerous parts to the design which is not complete; hopefully I've detailed enough to provide a clear understanding of the direction I'm taking the Colony Personal Browser. If you've made it this far please leave a comment. Do you think the design is good, think its got gaping holes, or just way too ambitious?

Argot Versioning - Part 3 - Remote Type Negotiation

A key concept of Argot is that it allows a client and server to perform type agreement dynamically. Introducing meta data versioning creates a number of issues when performing type agreement. The following goes into the details of Argot remote type negotiation and investigates the issues and some possible solutions.

To understand the problems of dynamic type negotiation, the fundamental concept of Argot data encoding must be understood. Argot meta data definitions are a reflection of how the data is encoded in communications. This is the opposite of Abstract Syntax Notation(ASN1) which defines the abstract meta data of a structure and then applies one of various encodings when the data is being written.

For instance, I'll use the Address data type as an example:

    (library.entry
      (library.definition u8ascii:”address” u8ascii:”1.0”)
      (sequence [
          (tag u8ascii:”street” (reference #u8ascii))
          (tag u8ascii:”suburb” (reference #u8ascii))
          (tag u8ascii:”state” (reference #u8ascii))
      ]))

In the above data definition, the address structure has three fields; street, suburb and state which are all defined as ASCII strings with a maximum length of 256 characters (u8ascii). Defining an instance of this in Argot would be:

  (address street:”PO Box 4591” suburb:”Melbourne” state:”Victoria”)

If this instance was to be serialised for communications it would look as follows:

    0x0B “PO Box 4591” 0x09 “Melbourne” 0x08 “Victoria”
  (note strings have not been changed to hex to ease readability)

Referring back to the address meta data you can see that this encoding shows a sequence of three strings. The u8ascii type uses a unsigned 8-bit byte to specify the length of each string before the data. Other than the length of each string before the string data there is no other meta data embedded into to encoding. This format has a number of consequences for how Argot must read data from the stream. The most important requirement is that Argot must know exactly how many fields each structure contains and what data is coming next. The advantage is that Argot is able to use a very compact data format with little to no meta data being utilised in the data stream. The disadvantage is that the exact structure of the data must be known before it can be read. This requires a client/server to both agree on the data types being used for communication.

Remote Dynamic Data Type Agreement

When communicating between client and server using Argot, information such as the Address instance above are identified using a 16bit identifier. This identifier is assigned dynamically to allow the communication channel to dynamically discover the data types it can communicate between client and server (as was discussed in the Versioning Part 1). In Argot's current form this negotiation is quite simple. It involves the following transactions.

• Meta dictionary check – There is an initial call to the server which checks if the core meta dictionary data types are the same. These data types are assigned a common set of identifiers that both client and server must adhere. This operates as a bootstrap mechanism for other types to be defined. This boot strap mechanism allows the meta data to be expanded and include new concepts for how to describe the data being transferred that were not previously developed in the core of Argot's meta data.
  
  
• Resolve Identifier – When a client needs to send a data type for the first time (that is not in the meta dictionary list of types) it sends a “resolve” message to the server. This message has the type's name and structure. The server receives the message, finds the data type and decides if the data structures match. If they match, an identifier is assigned for the channel for that type; if they don't match, an error is returned to the client. If an agreement can not be found, then the client will receive an error and the message being sent must be aborted.
  
  
• Reserve – In some circumstances a data type will have a cyclic reference to itself. Before the data type structure can be resolved using the above call the client must assign an identifier so that its data structure meta data can be written. The reserve call sends a message with a type name to the server. If the server has a data type with the same name it assigns an identifier to the client. If the server does not have the specific data type name then an error is returned to the client.
  
  
• Resolve Reverse – When a server is responding to a request it may wish to send a client a data structure that has not been resolved by the client. Argot uses asynchronous request/response semantics for all calls. This means that the server is unable to initiate a request to the client to resolve the data structure. In this situation the server assigns an identifier and sends the data to the client. When the client reads an identifier it doesn't recognise, it makes a “resolve reverse” call to the server with the identifier. The server responds with the name and the data type meta data assigned to the identifier. If the client finds it has the same data structure it is able to continue reading the data. If the client does not find a match it must abort reading the data as it does not understand the data received.
  

These four calls work well in Argot without versioning. The client and server are able to check for each and every type if the structure's match. This includes the data type meta data.

Adding versioning into Argot requires a location to be used instead of a name for all of the above calls. To a large extent, this is all that is required. However, there is a problem at the protocol level centred around the concept of “resolve reverse”. As stated above, currently the server has one version of each data type. When responding with a previously unused data type it is able to respond with the single version of the data structure. The client either reads the data or doesn't. In a situation where the server has multiple versions of the data structure it needs to know which version to send to the client. In a protocol that uses asynchronous request/reply semantics, the server is unable to initiate request to the client to ask which version to use.

Here's some ideas that were explored to solve this issue:

• Place holder – The server could return a place holder for the data instead of the actual data. This would require that the client must find this place holder in the data stream and send a message to the server.
  
  This is not suitable as it puts a burden on the application to keep the instance data around to be encoded as required by the client. It also makes the Argot streaming interfaces very complex.
  
  
• Second Channel – Require that the client hold open a second channel to the server. This allows the server to initiate requests such as this to the client. The client could close both channels when communications has completed.
  
  This is not suitable as TCP sessions are a scarce resource. Keeping open a communications channel for the chance of communications is not appropriate.
  
  
• Pause Stream – Stop the current response and return a message in the stream asking the client to resolve the version required. The client would resolve the version and then return a message to the server asking it to continue with the selected version.
  
  This is not suitable as the client may be in the middle of reading any other data type. It may not be in a position to find the message in the stream.
  

A few possible workable solutions are:

• Chunked Stream – Require that a response stream is broken up into chunks. The server fills up a chunk before sending the response to the client. If the server finds a data type that needs a version selected, the server can initiate a request to the client asking which version to use. As the stream is chunked the client is able to receive requests from the server interleaved with the response stream.
  
  The interesting part of this solution is it changes the underlying request/response semantics and opens up the stream to be a bidirectional group of channels. This aligns well with the asynchronous request/reply concept already stated in BORED (Binary Object REst Distributed system). Using a chunked stream is bringing the concepts of TCP up a layer to allow multiple communications to occur on the same channel. Allowing the server to initiate requests also creates a new set of opportunities and challenges. The chunked stream also has some similarities to SCTP, the protocol used in VOIP systems.
  
  
• Pre-fetch – Require that the client know the type of data structures that will be returned by a request. The client must send a request to the server with all the data types that it could receive in the response.
  
  Ideally the client would send a group of data types to the server for data type and version negotiation. An issue with this is that some data types may need to be resolved before the group can be sent. To resolve this, the client can use a set group of reserved identifiers for the purpose of performing type resolution.
  

Having two solutions only solves part of the problem. There are a few scenarios which should be catered for:

• Negotiated Versioning – The discussion above revolves around the concept that the client and server require to negotiate the version of data they should use. This scenario would be most effective where a client and server have a communication which persists over a series of calls. This is especially true in environments that have multiple servers and clients with data that is evolving over time.
  
  In negotiated versioning the server has a set of versions for each data type. The client is able to select which version it would like to communicate with for each type.
  
  
• Shared Versioning – Another possible scenario is that both the client and server agree to adhere to a version dictated by separate server. In this scenario the server is able to send any data to the client as long as it adheres to the shared version. This may be appropriate for organisations wishing to centralise the data dictionary for better management purposes. It is also likely that the server will contain multiple versions of data types. The client would select a version of the dictionary which would select the correct version of each data type; much like a version in traditional version control systems. This is required as clients and servers programmed for a particular version of a data type can not have the structure of that type changed unexpectedly.
  
  This scenario requires that the client or server select the server that will be used to select the data type version.
  
  
• Server Dictated – In some cases the server may only have a single version of each data type. In this scenario there is no point attempting to negotiate the data types. The client must have the specific versions of data required by the client. This is scenario that Argot currently uses for communications. In scenarios where deploying new servers are more expensive and clients talk to many servers this is most appropriate. The client must contain all data type versions required to communicate with each server.
  
  This case is also true for any form of message queueing or file base messages. The server will have no idea which version the client requires. If the server has multiple versions available then it will need to select the most appropriate version of each type for a given file. It is also true for embedded systems where the server will not have the resources or processing power to perform full negotiation of data types.
  
  
• Client Pre-Selected – In some cases its most appropriate for the client to pre-select the data type versions it expects to receive from the server. In this scenario the client sends a group of data types and versions that the server should use for communications. In this scenario the client is more expensive to deploy or a server needs to communicate with multiple versions of a client.
  
  This method fits into the pre-fetch method above.
  
  
• Version Controlled – In this scenario the stream includes a version selector for a group of data types. This is more in-line with how many systems currently operate. The disadvantage of this mechanism is that every data type must be pre-selected to be part of the group of a selected version.
  

If you can think of other methods of performing version agreement between servers please let me know so that I can add it to the list.

Catering for the various types of version management in a single protocol is not a simple task. The protocol redesign should also align with the BORED protocol discussed in previous posts. In addition to these requirements the final requirement is to build security into the protocol. These issues will be explored in a later post.

Argot Programming Model

Another piece in the versioning puzzle is how the programmer's API has changed for the most common functions of Argot. In the current system, Argot uses the concept of a TypeMap for mapping specific data types to a stream. Currently this does not include any type of version information. Some example code looks like:

    TypeMap map = new TypeMap( typeLibrary );
  map.map( 1, typeLibrary.getId(“u8”));
  map.map( 2, typeLibrary.getId(“u8ascii”));

In this scenario, the user is mapping a local identifiers to data types in the type library. To support versioning, the developer would need to specify which version of u8 and u8ascii they wanted from the TypeLibrary.

    TypeMap map = new TypeMap( typeLibrary );
  map.map( 1, typeLibrary.getId(“u8”, “1.0.0”));
  map.map( 2, typeLibrary.getId(“u8ascii”, “1.0.0”));

The problem with this is it re-introduces a specific version too early in the communications. The solution is the introduction of a TypeMapper interface which is passed into the TypeMap. The TypeMapper has the task of selecting which version of a data type is required at the time it is being used. The user simply creates the TypeMap with the required TypeMapper. The TypeMap initialises the TypeMapper which gives it a chance to map any required types. When a developer writes a data type that is not in the map, the TypeMapper is called which resolves which version of a type to use. Creating a type map now looks like:

    TypeMap map = new TypeMap( typeLibrary, new TypeMapperDynamic());

In this case a dynamic type map is being used to resolve the data types. It dynamically assigns any types required by the type map. A stream is created and written using:

    typeStream = new TypeOutputStream( stream, map );
  typeStream.writeObject( “address”, addressObject );

If we were to write a specific version, the API would change to:

    typeStream.writeObject( “address”, “1.2”, addressObject );

Once again this re-introduces a specific version too early. If the line above was on a server, any client or receiver of the data would be locked in to version 1.2 of the address. For this reason the first example is how objects should be written to the stream. This requires that the TypeMapper select the correct version of the address type.

Different type mappers can be created to deal with the various styles of type negotiation listed above. After an id has been mapped, any use of that name will tie directly to the specified version.

Meta Dictionary Update

Near the end of the last post I suggested a change to the meta dictionary which had the effect of only allowing a single version of any data type to be used on a stream. The requirement at the API to only use the name and not the version validates that this change would match the API. The meta dictionary has been updated to reflect this change. This changes very little in actual meta dictionary.

A consequence of this change is that an additional request/response pair is required for the traditional method of performing type agreement. The first time a client wishes to use a type it must send to the server the name of the type without specifying the version. The server maps a specific version to the type and returns the mapped identifier, the location of the definition and the definition structure. The client is able to check this against its local version. This method continues to use server dictated versioning and is a temporary solution until a more advanced protocol can be devised.

Another small update to the meta dictionary is naming. Currently types in Argot are defined using a short ascii string (e.g. “meta.abstract.map” ). In the meta dictionary this is defined as:

    (library.entry
      (library.definition u8ascii:"meta.name" meta.version:"1.3")
      (meta.reference #u8ascii))

While the string implies that the name has groupings, each name is simply a unique string. As the number of objects in the TypeLibrary increases it will become more difficult to find specific groups of types. The string also goes against a central concept of Argot; there's no need to define string based expressions for encodings. The solution is to change the meta.name to:

    (library.entry
      (library.definition
          meta.name:"meta.name_part" meta.version:"1.3")
      (meta.reference #u8utf8))

  (library.entry
      (library.definition
          meta.name:"meta.name" meta.version:"1.3")
      (meta.array
          (meta.reference #uint8)
          (meta.reference #meta.name_part”)))

This creates an array of name parts which is a more true representation of the name. This allows the TypeLibrary to build a hierarchy in the type library. For programmer simplicity a parser is still used for the text representation. However it would still be possible to write one of the entries above as:

    (library.entry
      (library.definition
          (meta.name [ u8utf8:“meta” u8utf8:”name_part ] )
          (meta.version major:1 minor:3))
      (meta.reference #u8utf8))

Another small change is from u8ascii to u8utf8. This allows a wider variety of languages to be used for names. In the future I'll introduce a meta.alias type as an extension to the meta dictionary. This will allow different languages (I.e. Spanish, Japanese, etc.) to define their own names for data types while still keeping compatibility.

Object Relationships

Another issue to add to the list. How to you deal with the relationship between data types and data objects across multiple versions. The whole point of Argot is to create a simple API for making it easy to read/write data to/from data streams. In essence to communicate knowledge between systems.

When binding a Java class to the TypeLibrary, is the Name or the data type definition version used? If the same class can be used for all Definitions then binding to the name is appropriate. If different classes are required between versions then binding to the definition is required. All definitions should define a common interface or super class. If this is not the case then the developer must be very careful not to create data streams that intermix objects as class cast exceptions will be likely. This is another area which is not fully developed and will need to be explored. However, relative to the versioning base meta data changes, this is a small task.

Conclusion

The Argot library now supports versioning, however, there's still some loose ends to tidy up. Future posts will explore the some of these loose ends. Version 1.3.0 which includes versioning is currently being cleaned up and will be released in the next week or two.

Argot Versioning - Part 2 - Meta Data Naming

In this post I'll introduce the solution implemented for meta data versioning in Argot. It builds on the last post which introduced some of the versioning issues. Some light reading for getting the brain in gear after easter.

During the development of the versioning feature, a very important aspect of the system has been modified and updated. I found that every type definition needs more than a simple ascii string to define its name. Instead of a name, a location in the the type library is defined. To explain further, it's best to understand some background information and what this means for Argot.

To recap on the last post, performing type negotiation between peers (client and server) or application and file in the past requires each data type definition have a unique name. This has caused an issue with various aspects of meta data requiring a name where it has not been essential. This is because the basis of Argot is a single table which contains an identifier, name and definition.

Adding versioning into the meta data causes the single table to be broken up into multiple levels. Each name in the table may have multiple definitions. The small table example given in the last post now expands to a much larger table as shown in the table below.



Another example in Argot without versioning is that of abstract data types. These required multiple named definitions. A short example is:

 meta.definition: meta.abstract();
 meta.definition#basic: meta.map( #meta.definition, #meta.basic );
 meta.definition#map: meta.map( #meta.definition, #meta.map );

The three definitions is actually trying to represent the following:



This diagram represents three levels to the data type structure. The first entry defines the name (meta.definition). The second entry defines version 1.0 as being an abstract type. The third and fourth entries are a relation to the version 1.0 definition and map the abstract type to other types.

Using the same naming mechanism to flatten this into a single table useful for Argot creates a group of ugly name strings:

 id:10, name:”meta.definition” - meta.name;
 id:11, name: meta.definition#v1.0  - meta.abstract;
 id 12, name: meta.definition#meta.basic#v1.0 - meta.abstract.map #meta.basic;
 id 13, name: meta.definition#meta.map#v1.0 - meta.abstract.map #meta.map;

The solution to this is to replace each name with a location. The location is an abstract type that initially has three concrete location types. The first location includes the name, the second is a version definition and includes the id of the name location and the version information. The third location is a relation type and includes the id of a versioned definition( eg 11 in the above list) and a tag. The tag is a unique string used to uniquely identify the location. As in the flat table version of Argot where every name must be unique, a location must also be unique. It must be possible to find any definition using just its location data.

The separation of location from the definition is the key concept in Argot with versioning. The location being an abstract type also means that it can be extended to include any type of location specifier. The location specifier replaces the name and provides a flexible method to specify where to place a definition in the meta data library.

An interesting aspect of the above is that there is often more information being used to specify where the data belongs than the actual data. The abstract type "meta.definition" and mapping data definitions now look like:

 // 1. define the name.
 (library.entry 
  (dictionary.name:”meta.definition”) 
  (meta.identity) )

 // 2. define version 1.0 as abstract.
 (library.entry 
  (dictionary.definition name:”meta.definition” version:”1.0”)
  (meta.abstract [])

 // 3. map meta.basic to the abstract type.
 (library.entry 
  (dictionary.definition name:”meta.definition” version:”1.0”)
  (meta.abstract.map #meta.basic))

 // 4. map meta.abstract.map to the abstract type.
 (library.entry 
  (definition name:”meta.definition” version:”1.0”)
  (meta.abstract.map #meta.abstract.map)

Each entry in the above is in two parts, the location and the definition. This separation has also had other beneficial flow on effects. In the previous versions of Argot, information in the name string had to be replicated in the definition. In effect the definition was previously being used to specify both location and definition information. By using a data structure in the location, this is no longer required. An example of this is the “meta.abstract.map” definition which previously included both the abstract target and the mapping type. This now only includes the mapping.

The location information provides a mechanism that allows very flexible data structures to be defined in the data type library. This can be extended to define methods signatures or other methods of defining protocol semenatics. In effect it allows the type library to define a complex directed graph while still providing a flat one dimensional table structure so that each individual definition can be found.

Dictionary Text Format

An obvious change in the example above is that the syntax used to define a data type has also changed. The syntax is loosely based on LISP and provides a more flexible way of encoding the meta data in a text format.

Each parenthesis starts with the name of the data type. All subsequence elements is the data for that type. eg.

 (library.definition name:”empty” version:”1.3)

This is an instance of the “library.definition”(v1.3) data type. The library.definition is defined as follows:

 (library.entry
  (library.definition name:”library.definition” version:”1.3”)
  (meta.sequence [
   (meta.tag name:“name” (meta.reference #meta.name))
   (meta.tag name:“version” (meta.reference #meta.version))
  ]))

This shows that each list shown in parenthesis is actually a strict data structure.

Also in the example is how to include simple type data. “name” and “version” are the names of the fields in the library.definition structure. Field names can be specified for both simple types and data structure. The values for each follow the colon. i.e.

“field name”:”value” // not currently implemented
or
“field type”:”value”
or
“field name”:(“data structure” … ) // not currently implemented

For all value types the “field type” must provide a parser capable of parsing the value into an object used internally. In some cases a parser may be provided to parse a string into a complex internal structure. This is currently used for the meta.version type which uses MAJOR.MINOR string type.

The only other form are arrays. Arrays are specified using square brackets. e.g.

[ element1 element2 element3 ]
or
“field name”:[ element1 element2 element3 ] // not currently implemented


Meta Dictionary

The following is the full meta dictionary in its pre-compiled form. Each and every data type and structure used is defined in the meta dictionary. This provides the self referencing base from which all elements are defined. It does not attempt to define all basic data types. It only attempts to define those data types required as part of the meta dictionary. The data structures in the meta dictionary are used later to define all other common data types in the common dictionary.

You might want to skip the meta dictionary definition unless you really want to give the brain a work out.

// 0. empty
(library.entry 
 (library.definition u8ascii:"empty" meta.version:"1.3") 
 (meta.fixed_width uint16:0
  [ (meta.fixed_width.attribute.size uint16:0) ]))

// 1. uint8
(library.entry
 (library.definition u8ascii:"uint8" meta.version:"1.3")
 (meta.fixed_width uint16:8 
  [ (meta.fixed_width.attribute.size uint16:8)
      (meta.fixed_width.attribute.integer)
      (meta.fixed_width.attribute.unsigned)
      (meta.fixed_width.attribute.bigendian) ] ))

// 2. uint16
(library.entry
 (library.definition u8ascii:"uint16" meta.version:"1.3")
 (meta.fixed_width uint16:16
 [ (meta.fixed_width.attribute.size uint16:16)
   (meta.fixed_width.attribute.integer)
   (meta.fixed_width.attribute.unsigned)
   (meta.fixed_width.attribute.bigendian) ] ))

// 3. meta.id
(library.entry 
 (library.definition u8ascii:"meta.id" meta.version:"1.3")
 (meta.reference #uint16))

// 4. meta.abstract.map
(library.entry 
 (library.definition u8ascii:"meta.abstract.map" meta.version:"1.3") 
 (meta.sequence [
  (meta.tag u8ascii:"id" (meta.reference #meta.id))
 ]))

// 5. meta.abstract
(library.entry
 (library.definition u8ascii:"meta.abstract" meta.version:"1.3")
 (meta.sequence [
  (meta.array 
   (meta.reference #uint8)
   (meta.reference #meta.abstract.map))]))

// 6. u8ascii
(library.entry
 (library.definition u8ascii:"u8ascii" meta.version:"1.3") 
 (meta.encoding 
  (meta.array 
   (meta.reference #uint8) 
   (meta.reference #uint8))
  u8ascii:"ISO646-US")) 

// 7. meta.name
(library.entry
 (library.definition u8ascii:"meta.name" meta.version:"1.3")
 (meta.reference #u8ascii))

// 8. meta.version
(library.entry 
 (library.definition u8ascii:"meta.version" meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:”major” (meta.reference #uint8))
  (meta.tag u8ascii:”minor” (meta.reference #uint8))
 ]))


// 9. meta.definition
(library.entry
 (library.definition u8ascii:"meta.definition" meta.version:"1.3")
 (meta.abstract [
  (meta.abstract.map #meta.fixed_width)
  (meta.abstract.map #meta.abstract)
  (meta.abstract.map #meta.abstract.map)
  (meta.abstract.map #meta.expression)
  (meta.abstract.map #meta.identity)
 ]))

// 10. meta.identity
(library.entry
 (library.definition u8ascii:"meta.identity" meta.version:"1.3")
 (meta.sequence [
 ]))

// 11. meta.expression
(library.entry 
 (library.definition u8ascii:"meta.expression" meta.version:"1.3")
 (meta.abstract [
  (meta.abstract.map #meta.reference)
  (meta.abstract.map #meta.tag)
  (meta.abstract.map #meta.sequence)
  (meta.abstract.map #meta.array)
  (meta.abstract.map #meta.envelop)
  (meta.abstract.map #meta.encoding)
 ]))

// 12. meta.reference
(library.entry 
 (library.definition u8ascii:"meta.reference" meta.version:"1.3")
 (meta.sequence [(meta.reference #meta.id)]))

// 13. meta.tag
(library.entry
 (library.definition u8ascii:"meta.tag" meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"name" 
   (meta.reference #u8ascii))
  (meta.tag u8ascii:"data" 
   (meta.reference #meta.expression))]))

// 14. meta.sequence
(library.entry 
 (library.definition u8ascii:"meta.sequence" meta.version:"1.3")
 (meta.array 
  (meta.reference #uint8)
  (meta.reference #meta.expression)))

// 15. meta.array
(library.entry
 (library.definition u8ascii:"meta.array" meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"size" (meta.reference #meta.expression))
  (meta.tag u8ascii:"data" (meta.reference #meta.expression))]))

// 16. meta.envelop
(library.entry
 (library.definition u8ascii:"meta.envelop" meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"size" 
   (meta.reference #meta.expression))
  (meta.tag u8ascii:"type" 
   (meta.reference #meta.expression)) ]))


// 17. meta.encoding
(library.entry 
 (library.definition u8ascii:"meta.encoding" meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"data" (meta.reference #meta.expression))
  (meta.tag u8ascii:"encoding" (meta.reference #u8ascii))]))

// 18. meta.fixed_width
(library.entry 
 (library.definition u8ascii:"meta.fixed_width" meta.version:"1.3") 
 (meta.sequence [
  (meta.tag u8ascii:"size" (meta.reference #uint16))
     (meta.tag u8ascii:"flags" 
   (meta.array 
    (meta.reference #uint8)
    (meta.reference #meta.fixed_width.attribute)))]))

// 19. meta.fixed_width.attribute
(library.entry
 (library.definition 
  u8ascii:"meta.fixed_width.attribute" meta.version:"1.3")
 (meta.abstract [
  (meta.abstract.map #meta.fixed_width.attribute.size)
  (meta.abstract.map #meta.fixed_width.attribute.integer)
  (meta.abstract.map #meta.fixed_width.attribute.unsigned)
  (meta.abstract.map #meta.fixed_width.attribute.bigendian)
 ]))

// 20. meta.fixed_width.attribute.size
(library.entry
 (library.definition 
  u8ascii:"meta.fixed_width.attribute.size" meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"size" (meta.reference #uint16))
 ]))
 
// 21. meta.fixed_width.attribute.integer
(library.entry
 (library.definition 
  u8ascii:"meta.fixed_width.attribute.integer"
  meta.version:"1.3")
 (meta.sequence []))

// 22. meta.fixed_width.attribute.unsigned
(library.entry 
 (library.definition 
  u8ascii:"meta.fixed_width.attribute.unsigned" 
  meta.version:"1.3") 
 (meta.sequence []))

// 23. meta.fixed_width.attribute.bigendian
(library.entry 
 (library.definition 
  u8ascii:"meta.fixed_width.attribute.bigendian" 
  meta.version:"1.3")
 (meta.sequence[]))


// 24. dictionary.name
(library.entry
 (library.definition u8ascii:"dictionary.name" meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"name" (meta.reference #meta.name))
 ]))

// 25. dictionary.definition
(library.entry
 (library.definition u8ascii:"dictionary.definition"      meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"id" (meta.reference #meta.id))
  (meta.tag u8ascii:"version" (meta.reference #meta.version))
 ]))

// 26. dictionary.relation
(library.entry
 (library.definition u8ascii:"dictionary.relation"    
  meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"id" (meta.reference #meta.id))
 ]))


// 27. dictionary.location
(library.entry
 (library.definition u8ascii:"dictionary.location" 
  meta.version:"1.3")
 (meta.abstract [
  (meta.abstract.map #dictionary.name)
  (meta.abstract.map #dictionary.definition)
  (meta.abstract.map #dictionary.relation)
 ]))

// 28. dictionary.definition.envelop
(library.entry
 (library.definition 
  u8ascii:"meta.definition.envelop" 
  meta.version:"1.3")
 (meta.envelop 
  (meta.reference #uint16)
  (meta.reference #meta.definition)))

// 29. dictionary.entry
(library.entry 
 (library.definition u8ascii:"dictionary.entry" meta.version:"1.3")
 (meta.sequence [
  (meta.tag u8ascii:"id" 
   (meta.reference #meta.id))
  (meta.tag u8ascii:"location" 
   (meta.reference #dictionary.location))
  (meta.tag u8ascii:"definition" 
   (meta.reference #meta.definition.envelop))]))

// 30. dictionary.entry.list
(library.entry
 (library.definition u8ascii:"dictionary.entry.list" 
  meta.version:"1.3")
 (meta.array 
  (meta.reference #uint16)
  (meta.reference #dictionary.entry )))

Library types

These types are only used for the pre-compiled definitions and are used by the compiler. They are kept separate from the meta dictionary. These are required so that a user does not need to define identifiers for each type and keep track of which entry is defined by which identifier.

// library.entry
(library.entry
 (library.definition name:”library.entry” meta.version:”1.3”)
 (meta.sequence [
  (meta.tag “location” (meta.reference #library.location)
  (meta.tag “definition” (meta.reference #meta.definition)
 ]))

// library.location
(library.entry
 (library.definition name:”library.location” meta.version:”1.3”)
 (meta.abstract [
  (meta.abstract.map #library.definition)
  ]))

// library.definition
(library.entry
 (library.definition name:”library.definition” meta.version:”1.3”)
 (meta.sequence [
  (meta.tag “name” (meta.reference #meta.name))
  (meta.tag “version” (meta.reference #meta.version))
 ]))

Multiple Versions Per Stream

Each of the entries in the meta dictionary is compiled into two dictionary entries. For example the “empty” data type is defined by the following two dictionary entries:

 (dictionary.entry
  meta.id:1
  (dictionary.name name:”empty”) 
  (meta.identity))

 (dictionary.entry 
  meta.id:2
  (dictionary.definition #empty meta.version:”1.3”)
  (meta.fixed_width size:0
   [ (meta.fixed_width.attribute.size size:0) ]))

This fits the versioning model of Argot. The first entry simply defines the name (“empty), while the second entry defines the definition of version 1.3 of the “empty” type. This mimics the internal representation of the type library. A question I am yet to resolve; should this be the external representation? Another solution for the external representation combines the two entries:

 (dictionary.entry 
  meta.id:2
  (dictionary.definition u8ascii:”empty” meta.version:”1.3”)
  (meta.fixed_width size:0
   [ (meta.fixed_width.attribute.size size:0) ]))

The advantage of this is that it reduces the data size for the dictionary. A consequence of this is that the meta identifier (meta.id) is the same for both the name and the specific meta data version (in this case 1.3). Therefore only the one version of a data type can be used in any individual communication or stream. This is possibly an advantage, as the constraint will create an easier to debug and program communications environment. It also allows a simpler API to be developed which only needs to map each named type to a single version. The disadvantage is that it reduces the flexibility of the communications environment; there may be situations where multiple versions of the same type need to be communicated in the one stream.

Nearly Full Circle

Another adaption to the above is to remove the version from the definition. e.g.

 (dictionary.entry 
  meta.id:2
  (dictionary.name u8ascii:”empty”)
  (meta.fixed_width size:0
   [ (meta.fixed_width.attribute.size size:0) ]))

The removal of the version information requires that each definition is used to create a unique signature. The signature becomes the version data used to match particular versions. This method is very close to the original method of defining data, however, the location instead of name is still required. The location type allows the relation location type to be used for abstract types and other types that are defined using multiple entries. This disadvantage of removing the version data is that it requires a more complex library and doesn't provide any form of ordering to be performed between versions. For this reason it won't be used.


Conclusion

The solution implemented for versioning meta data in Argot provides a new and innovative approach to this difficult problem. The concept of using a location in a directed graph allows any graph to be built and partially compared. In the next post I'll explore the area of remote data type negotiation and show how versioning adds new complexities.

Argot Versioning - Part 1 - Meta Data Type Versioning

Data type meta data versioning in communication is one of the more (if not most) complex areas of distributed computing. It's also an area that I've managed to avoid for quite some time. That is until someone sent me an email asking about versioning in Argot. This has kicked off a thought process which has resulted in around six months of investigation, a couple failed attempts, and introducing a long overdue feature in Argot. The final result solves the problem of versioning in a new and unique way.

In the next series of posts I'm going to discuss some of the issues of versioning and how versioning has been implemented in Argot. Be warned that the posts are probably going to be long and complex. Along the way, I'll demonstrate the new Argot meta dictionary which will become the basis for versioning meta (type) data in Argot.

Background

In Argot, before now, I've taken the view that a client and server have a single version of each data type. Each data type has a name, a structure definition and a unique identifier. When an Argot connection is established the client and server compare the names and data structures for each type. If the data structure of any type is different between client and server then the system is unable to communicate that data type.

The image above demonstrates how a client and server create a shared table which defines the set of data types that they can use to communicate. The name and definition of each data type must be the same on both client and server for an entry to be added to the shared table. The client and server assign a unique internal identifier which may differ for their own data type tables; each data type in the shared table has a unique identifier that is agreed between client and server.

For my purposes this method of having a single data type has worked fine. In my small environments I can update the client and server at the same time. However, versioning is a necessary requirement for many systems. You can't always upgrade all clients after a server has been updated. This means a single server must be able to support multiple versions on the client. In a similar way, you can't always upgrade all servers, requiring a client to support multiple versions of data structures. In situations where both clients and servers can not be upgraded then both must have multiple versions of data types. Therefore Argot needs to be modified to allow a data type name to have multiple definitions or versions.

The Issue of Names

The development of Argot to some extent has always been based on a language dictionary. The idea is that each and every data type definition can be taken individually (much like a single word can be found in a dictionary) and used in any data dictionary (schema). The language dictionary is once again the premise for how versioning should be handled by Argot. A standard language dictionary defines various aspects of a word's definition. Each word will have its pronunciation, phonetic spelling, various ways the word is used and possibly its etymology.

Compare this to an example using Argot's (version 1.2) definition:

address:   
      meta.sequence([  
         meta.reference( #u8ascii, "street"),
  meta.reference( #u8ascii, "suburb"),
  meta.reference( #u8ascii, "state" )  
 ]);

Argot provides a very basic format to create definitions. It has two parts: the name ("address" in the example above) and the definition. Internally this is also assigned a unique identifier. This simplicity is a double edged sword. The consequence is that every definition must have a name. However, there are many cases where a name is not required for Argot definitions.

The other important aspect of Argot is that each statement or definition must stand alone. This is required so that a client and server can compare each part of a types definition. This means that a single concept may be defined using multiple statements. This is the case for abstract data types. For example:

 meta.definition: meta.abstract();
 meta.definition#basic: meta.map( #meta.definition, #meta.basic );
 meta.definition#map: meta.map( #meta.definition, #meta.map );

In this case meta.definition is defined as an abstract data type. The meta.basic and meta.map are then mapped to the abstract type using separate definitions. This requires that Argot define fake names like "meta.definition#basic" so that each definition can be found in the data type tables.

Introducing versioning offers an opportunity to modify the way data types are defined to create a model which is closer to a language dictionary.

An interesting aspect of basing versioning on a language dictionary is that each version of a data type may be completely different. A single dictionary might define an address as:

 address
  version:"1.0" :
       sequence( [
    reference( #u8ascii, "street" )
    reference( #u8ascii, "suburb state" )
       ] );
  version"2.0" :
       sequence( [
    sequence( [
     reference( #u8utf8, "street number" )
     reference( #u8utf8, "street name" )
     reference( #u8utf8, "street type" )
   ])
   reference( #street, "street" )
   reference( #u8utf8, "suburb" )
   reference( #u8utf8, "city" )
   reference( #u8utf8, "state" )
      ] );

This is considerably different to how many other object serialization systems work. For instance, in ProtocolBuffers a label is assigned to each field in a definition. New versions consist of adding new optional elements to the definition. In effect this means a definition can not change radically between versions. It also means that new versions must become a hybrid data structure of both old and new, moving the strict rules about the data structure into the program. The advantage of the language based model is that versions can be completely different and encode strict definitions at the protocol or file format level.

Versions in Structure Definitions

One of the first problems to be solved in introducing versioning is how to reference a data type with multiple versions in another definition. For example, in the following example the structure test refers to "foo" and "bar" version 1.0.

 test:
  version:"1.0":
       sequence( [
    reference(#foo, version:1.0)
    reference(#bar, version:1.0)
       ]))

However, there's a problem, when a data structure is defined and contains references to other data types it creates a brittle type system that is difficult to maintain. In the "test" definition there's a strict relationship of versioning between each sequence sub element. If foo was to be updated to "foo_1.1" the "test" type would also need to be updated. This causes a versioning ripple through every element that uses foo. Every element that was changed will also cause changes.

In the following example we try defining "test" using major and minor versions. Each reference can then specify the minimum version that is supported. The problem with this is that every definition requires too much data. The developer and schema designer will get lost in meta data versioning information.

"test" vMajor:1 vMinor:0:
 (reference #foo (minVersion major:1 minor:0) (maxVersion major:1 minor:99) );

Looking back at the dictionary model (ie real world dictionary book) that Argot was built upon, it is clear that every word definition does not refer directly to a specific version of each word used to define another. Returning back to the original concept of a data structure definition without version information:

"test" version:"1.0":
    sequence([
 reference( #foo)
 reference( #bar)
    ]));

In many cases the actual version of a referenced field is probably not important when defining the data type. As long as the server and client both agree on what version of a particular type they agree on then the data can be any format.

This model has the advantage that any change to "foo" does not cause a ripple through other data types. It also removes any barriers to what version of "foo" a client and server should use to communicate. This places additional burden on the programmer to ensure that all versions of "foo" that can be understood by the software are interchangeable through all parts of the application. Overall the advantages of not specifying a version is preferred over the other options, so it will be adopted for Argot versioning.

Using this model requires the ability to identify both the Name and definition as two separate references in the Argot Type system. When data structure's are being defined, any reference uses the Name identifier. When a data structure is being used in communication it uses a specific definition. This means that the name must have its own identifier and form part of the TypeLibrary.

Version Information Data

There are a few options as to how to encode the version information for a specified data structure. As far as Argot is concerned each version of a data type is a completely different type. From this point of view having a single integer value to represent the version is easiest. However, from a user point of view the version often consists of major, minor and patch levels.

Version options:

• Single Integer - Has the advantage that it aligns with the design of Argot. Each version is completely independent of the other. Ordering of versions can be easily maintained.
  
  
• Major, Minor, Patch Integers - Allows each data type to have multiple levels. Ordering of later versions can be easily maintained. The other advantage of this is it allows a mechanism for designers to make "compatible" changes to the protocol as part of minor revisions. A major revision will often signal a departure from previous logical designs and the previous system Object will no longer be a viable representation. A minor revision will signal the addition of a field or other minor change in which case the same system Object can be used. This method also allows a developer to keep track of the version of software that a definition was introduced.
  
  
• String - A generic string has the advantage that any versioning system that the developer produces can be handled. No ordering can be guaranteed unless an additional ordering function is supplied by the developer and bound to each data type.
  
  
• Abstract Type - An abstract type offers the most flexibility as the user is able to define the version using any method and mapping to the abstract type. This expands the meta dictionary and makes versions more difficult to compare.
  

To reduce complexity in the initial development a simple string was used. However, as the release version is developed this will migrate to a MAJOR and MINOR mechanism. The major and minor values are small unsigned integers. The use of major and minor releases becomes important to differentiate between new and older type versions. Later releases may eventually allow multiple tags to be assigned to specific versions providing a method of performing a version control across a group of data types.

This sets a couple of the core concepts of versioning in Argot. In the next post I'll introduce the key concept that makes versioning in Argot a possibility and demonstrate the Argot meta dictionary.

Part 7 - Uniform Interface Constraint

Its been a couple of weeks since the last BORED post. Something has gone wrong with my very old Thinkpad. It now only boots in safemode with 640x480 resolution; not the nicest environment to work. I'm holding out for another few weeks before hopefully getting hold of a new Mac. I'm still not sure if it will be an air, book, pro. The posts here will probably slow down until they're out.

For now, back to BORED. Today's post is probably the most interesting of the posts and highlights the real issue that BORED, Argot and every other protocol is really trying to solve. That is the movement of information and knowledge between client and server. This is very different to the simple task of moving data (ie bits and bytes). The problem of moving knowledge between applications is the central aspects of what draws me to this otherwise dull area of computer science.

I look at the current browsers, programming languages and enterprise systems and see a single underlying problem; we have very little understanding of how to move knowledge between systems. Solving this problem can lead to more fluidity of data with less work by programmers. This should also lead to better useability for the applications we build. There's a lot of work to do and probably a few books to be written in this area before it will be solved. BORED is an excersize in breaking out of the mold and seeing if a better approach can be found. Without further ado, lets get back to BORED!

The BORED protocol has now been tested against some of the challenging REST constraints. The next and probably the most difficult constraint to be tested is the Uniform Interface Constraint. This is the point where the request message data structures hit the target object and the mismatch between a hypermedia system and other types of interactions with servers is most obvious. As the aim of the BORED protocol is to bring find some alignment between REST and Object orientated systems, this is where things should get interesting.

Uniform Interface

The Uniform Interface constraint is one of the more interesting constraints of REST. It reduces all operations to a small set of file like operations, e.g. GET, POST, PUT, DELETE, HEAD, etc. In the case of BORED, however, I'm trying to bring together the concept of an Object Orientated system with that of a Hypermedia system in a sensible way. At this point it is a good time to review the BORED architectural model:

client --[request]-->Server-->Container-->Object ReceiverObject

client<--[response]--Server<--Container<--Object ReceiverObject

The BORED Remote Message Call(RMC) model encompasses all interface request data into the message data portion of the request. This is delivered to the Object Receiver, which uses this information to interact with the target Object. These interactions could involve any one of the following:

Object Receiver -------> Document/File
Object Receiver -------> Object Instance with public methods
Object Receiver -------> Data Collection
Object Receiver -------> Proxy Interface
Object Receiver -------> Etc...

It is also worth reviewing what Fielding has to say about the Uniform Interface Constraint:

"The central feature that distinguishes the REST architectural style from other network based styles is its emphasis on a uniform interface between components (Figure 5-6). By applying the software engineering principle of generality to the component interface, the overall system architecture is simplified and the visibility of interactions is improved. Implementations are decoupled from the services they provide, which encourages independent evolvability. The trade-off, though, is that a uniform interface degrades efficiency, since information is transferred in a standardized form rather than one which is specific to an application’s needs. The REST interface is designed to be efficient for large grain hypermedia data transfer, optimizing for the common case of the Web, but resulting in an interface that is not optimal for other forms of architectural interaction."

As stated, it is the Uniform Interface constraint that really sets the REST approach apart from many other systems. It is the simplicity of the uniform interface that makes the interactions between browser and web server so powerful.

Fielding continues with:

"In order to obtain a uniform interface, multiple architectural constraints are needed to guide the behaviour of components. REST is defined by four interface constraints: identification of resources; manipulation of resources through representations; self descriptive messages; and, hypermedia as the engine of application state. These constraints will be discussed in Section 5.2."

The Uniform Interface constraint therefore has multiple sub-constraints. Any diversion from these constraints will cause BORED to diverge from the REST approach. However, Fielding also states that the Uniform Interface constraint is a trade-off between degrading efficiency and providing an:

"efficient interface for large grain hypermedia data transfer, optimizing for the common case of the Web, but resulting in an interface that is not optimal for other forms of architectural interaction."

This trade-off is clearly shown in AJAX based applications. Application designers are forced to use the REST approach for all aspects of the client-server interactions. An AJAX based application downloads Javascript which often makes remote calls back to the web server. These AJAX calls are better suited to a solution which allows a program centric interaction with the server (note I'm being careful not to use the term RPC). The client may be attempting to return document fragments or even simple single string responses. In these situations the uniform interface constraint creates additional work for the developer and designer. These AJAX/Web 2.0 interactions would benefit from a stronger binding between client and server.

The AJAX/Web 2.0 example shows the trade-off that Fielding discusses in regards to REST. However, the trade-off has obviously served the Web Hypermedia system well to this point. Take for example the simplicity of:

http://www.livemedia.com.au/my_image.jpg

By entering a URL into a browser we imply the GET request, and the image is retrieved. The web's power is driven through this simplicity.

At this point it is worth doing a small detour into the realm of data contracts. Understanding the different types of data contracts that client/server systems use will provide a better set of tests to base the BORED protocol.

Data Contracts

The topic of data contracts is probably the most interesting aspect of distributed computing. This is where there is some agreement between client and server that after sending a specific set of data to a location will result in an agreed set of other data being returned. The contract can range from being implied, to being rigidly defined using procedure call semantics (as is the case in CORBA IDL). This philosophy around the area of data contracts changes with each new technology and fad.

The reason for this constant flux is that what is required changes for different purposes. If a user is involved then human cognition is the most important part of the contract. If the communication is purely between code on both client and server then as long as the client matches the server the contract can be implied. If the clients are many and varied and are using a 3rd party service then consistancy and an Interface Definition Language is desierable. If the client wishes to discover new interfaces then discoverability and associated IDL is a requirement. Finally, in some cases an IDL does not go far enough and a full and independent textual specification (eg RFC) is required.

Each of the methods of creating data contracts implies different requirements for the BORED protocol. The following is a simple breakdown of different contracts and some implications for the BORED protocol. There has probably been better and more thorough analysis of data contracts been done before; if you're aware of any, please let me know via comments.

Human Cognition Data Contract

The URL is probably the best example of providing human cognition to a data contract. By reading a URL a user is able to have a fairly good idea of what information will be returned. There is obviously skill in defining a good URL structure for any web site. However, the current URL also include request parameters which can modify the result of a particular page.

Take for example the following hyperthetical request:

http://www.livemedia.com.au/store_search.x?s=books&author=ryan&page=2

To an experienced web user the parameters of this request are obvious:

s=books (search the term “books”)
author=ryan (find books by the
author “ryan”)
page=2 (return the second page of results)

The contract between client and server regarding these parameters are loose and do not provide a formal contract between client and server. However, the need for this contract is reduced for most web pages as Javascript is usually used to construct the required URL. This reduces the usability of the server application for purposes outside the scope of the web page that uses it. To fix this issue many web sites that wish to offer services to external sites use additional web services based around SOAP technologies.

As part of the BORED protocol we've already stated a location requirement. At the sametime this ability for a user to modify a request URL parameters provides flexibility and gives the user a greater control over the information requested. To solve this in BORED a mechanism for the data contract for a URL to be made available to the browser. The user can then be presented with an interface to allow them to modify request parameters in a formal way.

Hypermedia Data Contract

The hypermedia data contract is associated with the REST design philosophy. The idea that each object in a system is identified using a simple URL. Links and references between objects are provided through hyperlinks. For instance instead of saying that a customer order contains product id A1234, a hyperlink is provided which gives a direct link to the component.

The final solution in a hypermedia system consists of many URLs with the interaction of each object in the system reduced to the REST uniform interface constraints. The client may request to be provided with different formats in responses depending on its requirements. Examples include HTML for user interaction, XML for programming language interaction and PDF for printing purposes. The REST approach requires that a client can specify the type of data it wishes to receive through its request parameters.

The actual data contract is deferred to the data representation returned by the specific request. In the case of XML, the contract can be specified using XML Schema or DTD. In a way, REST splits the data contract into two halves; The data representation using mime and the location of information through the publishing of URLs. REST does not currently have a method of bringing this information together in a way to make the data discoverable.

This type of solution puts minimal requirements on the BORED protocol. Infact, the hypermedia/REST method does its best to avoid the data contract issue by reducing all calls to a uniform interface and mime-types. The BORED protocol must allow the client to request the type of data it wishes to have returned in the GET verb. The response must support the ability to provide at least a mime type associated with the data returned.

Implied Data Contract

This form of contract is when a client is built specifically for a server. This is the case with most Web 2.0/AJAX based applications today. The data contract is not specified in any formal way and the specific data sent and received is at the full discretion of the developer. The developer must be aware of any changes with the data sent or received between client and server and ensure each section of the software is updated.

A large Web 2.0/AJAX based application is likely to have associated documentation that is put together by the team developing the product. This documentation is unlikely to provide a formal description of the data. Any changes to the product would require a separate task to update the associated documentation.

An important aspect of this type of communication is that the data format sent to/from the server is at the total discretion of the developer. A developer may return a simple string, some preformated HTML, some JSON data, or a new script.

This type of communication puts a requirement that the protocol should be open to any data type the developer wishes to sent between the client and server. To support this, the message data should allow at the very least a mime-type to be specified with the data. It may also be beneficial that additional meta data be provided separate to the data being supplied.

Strongly Typed Data Contract

This form of contract is defined using an Interface Definition Language(IDL) or Web Services Definition Language(WSDL). This is the most formal method of defining a contract between client and server. Associated with this type of formal contract is a complete communications stack (I.e. stubs and skeletons) which provides the middleware for the formal agreement.

In examples such as CORBA and Web Services there is a complete framework which surrounds strongly typed data contracts. As has already been shown numerous times over, a single framework is unlikely to provide all the facilities required by developers. Multiple Data Contract Languages (DCL's) may be required with more specific purposes to meet a developers requirements.

To support strongly typed data contracts the BORED protocol should support allow an interface to be associated with a specific location. In these cases the default GET uniform interface may provide an objects current state, however more specific interfaces can also be provided. The Data Contract Language associated with the object can be made available via a META call to the location.

Uniform Interface Constraint Reviewed

After reviewing the various types of contracts it is clear that the Uniform Interface constraint is just one of the many ways of building a data contract between client and server. However, the Uniform Interface constraint is an important element of the REST design; this is because as already suggested, its power is in its simplicity. eg

http://whereis.com

This URL links to a Web 2.0 application that uses heavy JavaScript and has many dynamic elements. In these cases the Uniform Interface constraint is defining a common naming mechanism that defines a link to a web application. The URL in effect is a bootstrap for the application.

The calls to the server after the application is loaded are hidden away from the user. If the user went hunting through the code for these AJAX calls they would be of little use alone. They are likely to require very specific parameters and return information that only the client application can interpret.

This concept of identifying the URL as a bootstrap has helped frame a few conclusions about the Uniform Interface constraint. In particular the requirement for the uniform interface constraint is only important for those locations that require direct user interaction or return a simple data representation. However, any data that needs to be transferred between the client application and server can use any other means that is suitable to the application. A developer may choose a mixture of implied and strongly typed interface contracts. It depends on the needs of the interactions that will occur between client and server.

Uniform Meta Interface

The discussion above points to the need for a uniform meta interface layer to be created. The purpose of this layer is to provide a mechanism to discover the meta data associated with the object being interacted. This meta data can describe the type of interactions that the object receiver will resposne. This layer should be flexible enough to cater for any type of data be sent and received. It should allow the interactions with the object to be described using meta data that is most appropriate for the service being delivered.

The danger of providing a single meta interface layer which could describe any number of interfaces and protocols is that the REST uniform interface is lost. I've already stated numerous times that the power of the REST architectural style is in the uniform interface. For this reason, a reduced number of verbs needs to be defined which allow the object to publish a basic set of known operations. This supports the main aim of the REST architectural style, but still allows flexibility in catering for other application protocols.

To support this concept of a uniform meta presentation layer, the object receiver must at a minimum respond to a request with a META verb. The response to this verb must be a description of the interfaces supported by the object. The meta data returned may include a set of REST style verbs such as GET, HEAD and DELETE.

This implies that for many interactions the client must make a call to the server to retrieve the meta data and then a second call to perform an operation. Additional calls may also be required to retrieve and bind interfaces depending on the method described in the meta data. However, it will be assumed that most locations will support atleast the GET verb with no parameters. This support will allow published URLs to be accessed without making multiple calls to the service.

Part 6 - Cache Constraint

The BORED protocol already meets the first two constraints of REST; client-server and stateless. We've also extended the client-server constraint to allow asynchronous client-server. The next REST constraint to meet is the Cache constraint.

Cache
Returning to Fielding's REST dissertation, we find:

"Cache constraints require that the data within a response to a request be implicitly or explicitly labelled as cacheable or non-cacheable. If a response is cacheable, then a client cache is given the right to reuse that response data for later, equivalent requests."

In the BORED protocol there's an additional requirement to this, which relates to the stateless requirement. To label a response as cacheable or non-cacheable requires that the request is uniquely identifiable. In BORED, the stateless request data is broken into two parts; the location and the message data. To satisfy this constraint a proxy server or client must identify the location and the request data as a single object and match this against the response data. As the request message data is binary the simplest solution is for a client or proxy server to keep a hash on the message data and location. To improve performance this hash value could be added to the request data to provide a key to a cache that will lower its overhead to calculate the key. It's important to add that the hash should only be based on the message data. This allows proxies to perform operations such as rerouteing of messages to new locations without needing to update the hash value.

To support the response aspect of the cache requirement, BORED includes cache information in the response header:

preamble - BORED
version
dictionary parts
available request slots
request identifier

response code
cache information

In the REST mismatches with HTTP Fielding writes:

"Differentiating Non-authoritative Responses
One weakness that still exists in HTTP is that there is no consistent mechanism for differentiating between authoritative responses, which are generated by the origin server in response to the current request, and non-authoritative responses that are obtained from an intermediary or cache without accessing the origin server. The distinction can be important for applications that require authoritative responses, such as the safety-critical information appliances used within the health industry, and for those times when an error response is returned and the client is left wondering whether the error was due to the origin or to some intermediary. Attempts to solve this using additional status codes did not succeed, since the authoritative nature is usually orthogonal to the response status.

HTTP/1.1 did add a mechanism to control cache behaviour such that the desire for an authoritative response can be indicated. The ’no-cache’ directive on a request message requires any cache to forward the request toward the origin server even if it has a cached copy of what is being requested. This allows a client to refresh a cached copy, which is known to be corrupted or stale. However, using this field on a regular basis interferes with the performance benefits of caching. A more general solution would be to require that responses be marked as non-authoritative whenever an action does not result in contacting the origin server. A Warning response header field was defined in HTTP/1.1 for this purpose (and others), but it has not been widely implemented in practice."

When the request message headers are developed in detail it will be important to include the ability to define a 'no-cache' directive. The cache information returned in the response should also indicate if the response is non-authoritative.

Location only constraint
At this point we add another new constraint to the system; the location only constraint. The location in each request should only include the location specific information. Request parameters must only be supplied in the message data. This constraint is designed to ensure the separation of the message data from the location data. This allows fast and easier routing of message data.

This constraint is a direct opposite of a common practise of encoding request parameters on to URI's in HTTP. For example:

http://www.livemedia.com.au/bookstore?author=ryan&page=1&list=10

In the BORED protocol the location must be separate from the message data.

(location bored://www.livemedia.com.au/bookstore)
(message author=ryan@page=1&list=10)

This constraint is designed to combine with the cache constraint to ensure message parameters are not confused with location data in cache systems. It also ensures that the required meta data to decode the message is included in the message meta data.

It is interesting to note that the cache constraint requires the stateless constraint to function. A cache must be able to deal with a whole message uniquely to operate correctly.

Part 5 - Stateless constraint

The initial constraints of REST were easy to provide a solution in BORED. In this post we tackle the stateless constraint. As BORED uses binary message data this constraint creates some unique challenges.

Stateless

The Stateless requirement is REST's second constraint. Fielding writes:

"We next add a constraint to the client-server interaction: communication must be stateless in nature, as in the client-stateless-server (CSS) style of Section 3.4.3 (Figure 5-3), such that each request from client to server must contain all of the information necessary to understand the request, and cannot take advantage of any stored context on the server. Session state is therefore kept entirely on the client."

To see the stateless requirement more clearly I'll review HTTP. Here's an example of a HTTP 1.1 request and response.

GET http://www.eienet.com.au/ HTTP 1.1
...

200 OK
...

The request encodes the full description of what the client is requesting in the URI and HTTP GET verb. To align with REST, BORED requires a similar location specifier. Let's assume a URI for now, however, to support embedded devices this will need to be more flexible.

To satisfy the stateless constraint, the following parts of BORED are required in the request:

prefix - BORED
version
request identifier
location - URI location or
other location type.
....
message
-- message meta data.
-- message - request data.
---- operation - GET,META,POST,METHOD,etc
---- message data

To meet the stateless requirement the BORED protocol includes the location and full request data.

In the case of a binary protocol an interesting addition is the inclusion of "message meta data". This is Argot specific however can be extended to any binary system that has a meta data definition. In the Argot case the meta data specifies the data structures of the data in the message.

The “message meta data” describes the message data, however, at this point there's no meta data to describe the actual request structure. To understand how BORED will solve this it is worth introducing the concept of an Argot Message Format. The Argot Message Format is designed to be completely self defining. Here's a short description from the Argot Programmer's guide.

---

Argot Message Files & Dictionaries

Argot message files are binary encoded files that provide the specification of their data with the data. An Argot file contains three parts; a meta dictionary, a data dictionary and the data.

The Argot Message Format allows the full specification of the data to be transferred with the data. This requires no external definition of the data. For an application to be able to read the file its type library must contain all the data types used in the file. A Type Map is generated from the data dictionary portion of the file to read the data. The general format of the file is:



The receiver of an Argot enabled file is able to read the dictionary and compare the data types of its own dictionary with that of the files. Once the types of the file dictionary have been matched with that of the application reading the file, the data can be read. This completely removes the need for a static common domain schema. Each application and file in effect contains its own schema.



This can be re-illustrated using the following venn diagram:



The process of reading a file involves:

1. Binary compare of meta dictionary map. The very first dictionary map of the meta dictionary is the core met dictionary. The only way to read this entry is by performing a binary compare. These are the base dictionary items used to describe new items. Please refer to the meta dictionary reference section for details of the core meta dictionary.
   
2. Build and read Meta dictionary. The rest of the meta dictionary is read and mapped between the application and file.
   
3. Read the Data dictionary. Using the Type Map produced from entries in the Meta Dictionary the Data dictionary is read. A Data dictionary type map is created based on the types identified.
   
4. Read the Data. Using the Data dictionary type map the actual data of the file is read.
   

The argot message format can be used anywhere that a data buffer can be transferred. In files, message oriented middleware, email, etc.

---

It would be easy to simply use the Argot Message Format as the full request structure to be delivered to the server. However, carrying the 'meta dictionary' with each and every request adds a lot of overhead. This would also hide the contents of the request data requiring a cache/proxy to read the meta dictionary, data dictionary and data before it can understand the request.

The solution used in BORED is to use the version information of the protocol as a monica for a data dictionary. When a server receives a request it uses the BORED protocol version to choose the corresponding data dictionary. This is like having the meta dictionary and data dictionary of the request at the start of every request. The request and response BORED message are themselves specified in this data dictionary.

The BORED request message however also requires a meta data section for times when the meta data for the request does not include data required by the object receiver. The message data dictionary expands on the request data dictionary to include elements required by the message.

Logically this looks as follows:

[ meta dictionary ] [ request data dictionary ]
---- [ request ... [ message [ message data dictionary] [data ]] ... ]

This allows the Request to logically contain the full meta dictionary, data dictionary, and data for the full BORED request in every message without the overhead of the full meta dictionary and data dictionary.

Using the above method has a drawback that the "request data dictionary" must define every aspect of the request message structure. This includes, security, cache information formats, header formats, and others. This creates an issue for very small devices that only support a subset of the request headers. A solution to this is to break the "request data dictionary" into parts. The client and server can then identify in their request and response the parts of the request data dictionary it supports. For simplicity the parts supported can be indicated via a bit-flag in the version part of the header. For instance, the version header could use three 8-bit flags. The first two would be the major and minor version with the third being the bit-flag for the parts of the request data dictionary supported.

Building on the last post, the request structure header now looks like for the request:

prefix – BORED
version
dictionary parts
request identifier
...

and response:

prefix - BORED
version
dictionary parts
available request slots
request identifier
...

Delivering the stateless constraint using a binary protocol has required developing a few tricks. In particular using the request version number as a key to meta dictionary and request data dictionary has allowed the solution to deliver a technically correct construct and still delivered the ability to reduce the amount of network traffic for each request/response. Using the bit-flag for specifying the parts request data dictionary supported has also allows the solution to scale from small devices to large full features systems.