Ivan Zuzak – blog

Blog moved to a new home at GitHub

Posted in Uncategorized by Ivan Zuzak on December 30, 2010

I’ve moved the blog to a new home at ivanzuzak.info which is hosted on GitHub using GitHub Pages. If you’ve been following my blog here on WordPress — you can stop and switch to the new one because I won’t be posting here any more. WordPress has been fun, but having control over my content sounds fun and GitHub is perfect for that. Thanks and all the best in 2011!

Tagged with: , ,

Pmrpc discovery and publish-subscribe support + systematization of cross-context browser communication systems

Posted in web development by Ivan Zuzak on June 15, 2010

This blog has moved to a new domain. You can view this post here.

The pmrpc cross-context browser communication library has grown from the last time I blogged about it. Last time I wrote about adding support for RPC-style communication for WebWorkers. Today I’ll first introduce two new features – dynamic discovery of remote procedures and a publish-subscribe communication model. Also, I’ll write about our systematization of existing systems for cross-context browser communication.

Discovery and publish subscribe

Up until now, pmrpc supported only remote procedure call as the communication model for inter-window and WebWorker communication. This model of communication is based on a client-server interaction where the client both a) has an object reference of the destination window/webworker context and b) knows the name of the remote procedure it wants to call. Here’s an example of a pmrpc RPC call:

// server [window object A] exposes a procedure in it's window
pmrpc.register( {
  publicProcedureName : "HelloPMRPC",
  procedure : function(printParam) { alert(printParam); }
} );
// client [window object B] calls the exposed procedure remotely, from its own window
pmrpc.call( {
  destination : window.frames["myIframe"],
  publicProcedureName : "HelloPMRPC",
  params : ["Hello World!"]
} );

This works well for cases when the client knows how to obtain a reference of the server context object. However, it turns out that in many real-world cases the client does not know which window/iframe/worker implements the procedure it wants to call. Mashups and widget portals are the best example. Imagine that both the client and the server are widgets on iGoogle. Although you may be in control of both the client and server iframe contents, in most cases you will not know the names of iframe elements which contain them and which are located on the top container page. And you need the values of the name attributes so that you can obtain a reference to the iframe object which you could then use to call postMessage.

This referencing problem was the motivation for implementing both the discovery and publish-subscribe features. Both features enable clients to call a remote procedure without a-priori knowing which destination object implements the  procedure. First, the publish-subscribe feature enables clients to perform a remote call without specifying the destination context object reference. If a publish-subscribe call is made, the pmrpc infrastructure will make a RPC call for the specified procedure to all window, iframe and WebWorker object it can find (thus simulating a “publish” event on a channel named by the remote procedure name). The API for using the publish-subscribe feature is simple; it reuses the pmrpc.call method and instead of passing the reference of the destination context object, a “publish” string parameter is passed instead:

// server [window object A] exposes a procedure in it's window
pmrpc.register( {
  publicProcedureName : "HelloPMRPC",
  procedure : function(printParam) { alert(printParam); }
} );
// client [window object B] calls the exposed procedure remotely, from its own window
pmrpc.call( {
  destination : "publish",
  publicProcedureName : "HelloPMRPC",
  params : ["Hello World!"]
} );

Under the hood, the publish-subscribe feature uses the discovery feature to discover all window, iframe and WebWorker context objects and retrieves their references. Also, the publish-subscribe feature may be used with other pmrpc features, like access control and retries.

The discovery feature enables dynamic discovery of procedures registered at remote contexts and their filtering using various optional criteria. Discovered procedures may be filtered by the destination window/iframe/webworker context object, destination context origin (specified as a regular expression) and destination procedure name (also specified as a regular expression). For each discovered procedure that wasn’t filtered out, the procedure name, origin of the destination context, the access control rules specified for the procedure and the destination context object. The feature is implemented as a new pmrpc methodpmrpc.discover. The method is asynchronous and accepts a parameter object which specifies the filters and a callback method which will be called when all registered procedures are discovered. Here’s an example:

// server [window object A] exposes a procedure in it's window
pmrpc.register( {
  publicProcedureName : "HelloPMRPC",
  procedure : function(printParam) { alert(printParam); }
} );
// client [window object B] discovers and calls the exposed procedure remotely, from its own window
var discoCallback = function(discoveredProcedures) {
  if (discoveredProcedures.length > 0) {
    pmrpc.call( {
      destination : discoveredProcedures[0].destination,
      publicProcedureName : "HelloPMRPC",
      params : ["Hello World!"]
    } );
pmrpc.discover( {
  destination : [window.frames["someiFrameName"], window.frames["myIframe"]],
  origin : ".*goodOrigin.*",
  publicProcedureName : ".*Hello.*",
  callback : discoCallback
} );

The implementation of the discover method is based on recursively visiting every iframe in the window.frames object starting from window.top:

  1. The method fetches a reference to the top window object (window.top), adds the reference to the result array and repeats the procedure for each nested iframe (by iterating through the window.frames object). This way, all iframes in the current window are visited. This process is depicted by the figure below.
  2. The method adds all local web worker objects to the result array.
  3. On each context object in the result array, a special pmrpc infrastructure method is invoked to obtain a list of registered procedures.
  4. The list of all obtained procedures is filtered by the specified criteria.
  5. The method invokes the callback, passing the list of discovered procedures as a parameter.

It’s not an ideal way of discovering remote contexts, but it’s better than nothing and works with the latest versions of all major browsers.

Systematization of cross-context browser communication systems

It’s evident that the WWW is becoming ever more componentized as almost every web site is a mashup, contains widgets or uses WebWorkers. As a result, there has been a increase in development of systems that enable communication between these different browser contexts – windows, iframes and WebWorkers. Pmrpc is only one example of such system and lots of other systems exist. However, as the number of these systems increases, the understanding of each systems’ capabilities and their comparison and evaluation becomes a more difficult task since no effort is being made to systematize this ecosystem.

This is why Marko and I are trying to create a systematization of existing cross-context communication systems. Our work on this systematization is in its early stages but is openly available to anyone at a pmrpc wiki page. At the moment we are trying to do the following:

  1. Make a list of all mechanisms for cross-context communication. We are including both in-browser mechanisms like the postMessage API, cookies and URL fragment, and 3rd party libraries like easyXDM and pmrpc.
  2. Define dimensions for classifying the gathered systems. Each dimension defines a system characteristic whereas values along each dimension define a set of possible design choices.  Currently, we have defined about a dozen dimension and are still thinking some through. Here are some examples: Type of system (browser built-in, pure client-side framework, server-mediated initialization framework, server-mediated communication framework), Transport mechanism (cookies, URL fragment, postMessage, …), Programming model (message-oriented, shared memory, RPC, publish-subscribe).
  3. Evaluate the listed systems according to the defined dimensions.

I’d love to hear what you think of the new pmrpc features and the systematization (see more examples at our API docs page)! Do you know of any cross-context communication systems that are not listed? Do you think some other system characteristic should be included in the classification? @izuzak

Why understanding REST is hard and what we should do about it – systematization, models and terminology for REST

Posted in web development by Ivan Zuzak on April 3, 2010

This blog has moved to a new domain. You can view this post here.

This is going to be another long post, so I’m using the introduction as an overview again.


This post is about understanding REST, the software architectural style behind the World Wide Web. My Ph.D. research, which I’ll write about some other time, pushed me on the road of REST and over the last year I’ve been reading lots of research papers, lots of blogslots of mailing lists, lots of tweets, lots of videos, wikis, books and IRC transcripts on REST and I’ve also recently started the This Week in REST wiki and blog. In other words, I’ve read almost everything I could find on REST. So, in the first part of this post I’ll write about several thoughts which stuck with me while researching REST:

  1. REST is and will continue to be important – it’s the foundation of the WWW and will be the foundation of its future stages and dimensions, like the Semantic Web and the Web of Things.
  2. Understanding REST is hard – the material on REST is fragmented and there is no clearly defined and systematized terminology or formal models used in discussions.
  3. We can and should fix the problem – there are enough motivated and smart people who can, through an open and collaborative process, create a better way of fully understanding REST.

In the second part of the post I’ll add to the pile of scattered fragments on understanding RESTful systems and describe an abstract model of a simplified REST user-agent using finite state machines. It’s a very basic model but serves the purpose of showing that it is both possible and useful to develop such formal models.

I hope this post will motivate people involved and interested in REST to contribute to a process for improving the understanding of REST. I was planing to write my thoughts as a paper for the First International Workshop on RESTful Design (WS-REST 2010) but didn’t have enough time. Nevertheless, I hope that people who are going to the workshop will talk about this issue. I’ll start a thread on the REST mailing list for following up on these ideas, so please comment there if you want to join the discussion (EDIT: thread is here).

Representational State Transfer and why it’s important

Although this post is about understanding the REpresentational State Transfer (REST) software architectural style, I’m not going to write another REST introduction. If you don’t know anything about REST, start from the wikipedia article and go from there.

What’s more important for this post is that REST is and will continue to be an essential part of the WWW.  First, through the Web’s core technologies, REST is responsible for most of the good properties of the current Web as a large-scale network-based system. Second, REST principles guide the development of both the next “major” generations of the Web, like the Semantic Web, the Web of Things and the Real-time Web, and the “minor” incremental changes of continuous evolution. Whereas the Semantic Web is about exposing and interlinking data on the WWW, the Web of Things is about connecting and exposing every physical thing to it and the Real-time Web is about real-time access to that data and things. Therefore, understanding the Web today, it’s future evolution and the whole technology jungle on top is based on the understanding of REST. Third, REST is important as a research subject on its own (although some people think it isn’t EDIT: Subbu elaborated on this misunderstanding below in the comment section) and will be analyzed and built upon to define new architectural styles.

Understanding REST

In my experience, understanding REST is far from easy. Of course, not everyone needs to understand REST and different people may want or need to understand REST at different levels and depths. However, despite it being a highly technical and academic concept not everyone should easily and fully grasp, I think it’s hard even for people aimed at – computer scientists, software engineers and web architects. After reading everything I could find on REST and the WWW, here are the problems I believe are responsible for this:

First, the REST master reference, Roy Fielding’s doctoral dissertation (and subsequent academic papers), is not completely suitable for fully understanding REST. For a doctoral dissertation, it’s severely lacks images explaining important concepts in the two key sections on REST. However, I bet everyone knows the single image in those sections almost by heart from how much it’s being repeated everywhere (for those who don’t, that’s the image on the right). The same could be said for the lack of (formal) models to describe specific properties, elements or views of REST, like HATEOAS. It’s as if the very clean and systematic approach of explaining software architecture and network-based architectural styles present in the first few chapters of his Ph.D. somehow vanished in those two sections. I am aware that there are no formal models for defining architectural styles as a whole (at least, there weren’t any then), but using models like state machines, Petri nets or process calculi to describe and explain parts of it would have definitely been of great benefit. It’s that absolute clarity of formal models that helps when you need to have a complete grasp of a concept. Furthermore, since models are abstract methods for explaining specific problems, people only need to research the underlying model after which the specific use of the model is unambiguously clear. And just to be clear, I think Roy’s dissertation is overall well written and has had a big impact on both the academia and industry, and I very well know that making everyone on the planet happy with a research paper is idiotic (looking for a good Ph.D. comics reference for this), but I still wish those two sections were a bit more thorough and polished.

Second, since Roy’s dissertation doesn’t have clear answers for all questions, people start discussions over understanding specific parts of REST all over the Web. Relevant discussions are scattered over many mailing lists, blogs, Twitter accounts, wikis, academic papers, videos and even IRC transcripts (see the introduction for links to some of these). More often than not these discussion are not focused on REST per se, but on anything related to the Web, various specifics concerning current and future standards and so on. And even more often, the discussions unnecessarily repeat previous discussions probably because the authors are unaware that the same discussion exists elsewhere or it’s that those are too difficult to search for. For example, the HATEOAS constraint is getting a lot of attention everywhere. So there’s lots of unorganized, scattered, duplicated and mutually disconnected fragments on REST and it’s unclear where and how to find the ones that answer your questions. No, Googling won’t help as often as you’d like, no, a series of unrelated articles piled up together won’t either (though Mike Amundsen’s tweets often will), and yes, a-big-unorganized-mess is kind of the point of the Web in general, however it’s not the optimal methodology for systematizing knowledge in a way required for understanding complex concepts like REST.

Third, these fragments on REST often use mutually different terminology, same terminology with different meaning or terminology which is not explicitly defined. My favorite example is the word “state” which is overloaded with unexplained and overlapping meaning. “Resource state”, “session state”, “control state”, “client state”, “server state”, “application state”, “transaction state”, “steady state” and “transient state” are some of the terms used and it is not completely clear what some of those mean, how they are related and which REST elements use, change and transfer them and when. For example, what is client state? Is it something related to the client connector type? Is it something stored on the client component? If so, then why is it not called user agent state? Is client state a union of session and application state? Or is application state a synonym for session state? Which states does a steady state relate to – client, session or application? Which entities may change session state and when? Yes, some of these concepts are completely clear, but some are not and different people use them with different meaning. Still, the most entertaining situations are the ones in which people write only “state” and then you have to figure out which state they are referring to. Yeah, good luck. And there are other both simple and complex examples, from stating that REST methods are to be executed over representations instead of resources and discussing if ATOM is RESTful or not to explaining just what is an application in terms of REST. Lastly, similar to Roy’s dissertation, these discussion rarely use diagrams or formal models to explain anything.

These problems confuse people trying to learn REST on a basic level and make it hard to discuss REST on deeper levels.

What we should do about it

Now, I’m not implying that these problems will cause the apocalypse if we don’t solve them, but I’d sure like it if they go away (and lots of other people would also). So here are my suggestions:

First, a mess is still better than nothing. Everyone should at least continue with thinking, writing and talking about REST and create more fragments. Chances are that those fragments will help a fair amount of people interested in REST. I definitely learned a lot from reading excellent blogs, tweets, papers and mailing list posts (Ryan’s explanation of REST to his wife is especially entertaining, besides educational).

Second, I think this problem can and should be solved collaboratively and openly, not by single person (not even Roy Fielding) or as just another academic paper. If there’s going to be agreement over terminology, it’s meaning, models, and other ways of making understanding REST easier, this agreement must be backed by people who have relevant REST experience and must be open to comments from everyone else. Furthermore, untangling the mess will require a lot of work. So, in my opinion, the REST community (whatever that is) should:

  1. Agree that there is a problem worth fixing – do we think that we can create a better, clearer and more systematized way of understanding and discussing about REST?
  2. Express interest in fixing it – is this something people want to contribute their time to?
  3. Agree on how to fix it – what should be our output (a RESTopedia, a document, video tutorials) and how would we all contribute to and moderate the process?
  4. Do it – spend time discussing and developing the output.
  5. Eat our dogfood – use whatever we produce. If we don’t use the terminology and models we agree upon, the the mess has only gotten bigger.

There are more than enough smart and motivated people with different backgrounds and experience with REST and RESTful HTTP to make this happen and it would have a big impact if done (even more if done right). I myself am not sure what the best output would be and how to achieve it, but would like it to be a hypermedia document available freely on the WWW, contain a systematized intersection of terminology, images, models and rationale that everyone agrees upon and focus on REST, using HTTP/URI/HTML only for examples (not the other way around).

Formal models for REST – a FSM model of a simplified RESTful user-agent

First off, developing useful formal models for understanding architectural styles isn’t easy since architectural styles are named sets of constraints commonly defined using natural language, rarely using formalisms. Also, models of REST concepts should include as many other REST concepts as possible – e.g. a model of HATEOAS should somehow clear up how application state, representations, methods, resources, steady states, transient states and other concepts all play their role in HATEOAS. It’s not easy to model this in a clear and simple way. Nevertheless, these kinds of models are especially important since they connect rather than disperse concepts.

Mike Amundsen’s and Jan Algermissen’s poking at REST steady states was very thought-provoking for me – excerpts pointing at something often ignored when discussing REST, but could be formally defined. This led me to try to model REST user-agents using some kind of a state machine. I soon found out there are lots of interesting papers on modeling hypermedia applications in general and even more blogs on modeling REST user-agents using some kind of state machine formalism (FSMs, statecharts, Petri nets or something else). Here are some of my thoughts after reading these papers and blogs:

  1. Most often, models are based on the “each page is a state and each link is a state transition” analogy. This is confusing and wrong for two reasons. First, it’s confusing since resource identifiers (e.g. URIs) are used to somehow address states of the application (e.g. an URI addresses a page which is mapped to a state) and state transitions (e.g. an URI is a link to a state which is mapped to a transition). In REST, resource identifiers are neither the single thing determining the state (e.g. two clients can perform the same GET request on a resource and get different representations which determine state) nor the single thing determining the transition between states (e.g. one client can perform a GET on a resource, the other could perform a PUT). Roy’s thesis defines what constitutes application state and thus when the state changes: “An application’s state is therefore defined by its pending requests, the topology of connected components, the active requests on those connectors, the data flow of representations in response to those requests, and the processing of those representations as they are received by the user agent.” Second, models often ignore some REST concepts like steady and transient states and are therefore useful only for understanding REST up to a certain level.
  2. State machines models are extremely simple to understand and a powerful tool for modeling. For example, state machines can also be used for model checking using temporal logic (e.g. check that from any state there is a link to the home page).

That’s why I really like Stu Charlton’s recent post on RESTful user-agents which identifies different types of state machines involved in user-agent operation. Stu nailed most of the things I wanted to write about, but nevertheless – below is my first attempt at a simplified finite state machine (FSM) model of RESTful user-agents. I won’t go into explaining FSMs in detail, so just check the Wikipedia article if you don’t know what FSMs are.

I’ll concentrate on recognizer-type finite state machines – specifically, nondeterministic FSMs (NFAs). NFAs are mathematically defined as a quintuple (Σ,S,s0,δ,F), where

  • Σ is the input alphabet (a finite, non-empty set of symbols).
  • S is a finite, non-empty set of states.
  • s0 is an initial state, an element of S.
  • δ is the state-transition function: δ : S x Σ → P(S) where P(S) the power set S.
  • F is the set of final states, a subset of S.

The operation of an NFA automaton, as state machines are sometimes called, is described as follows. The NFA starts from the initial state and then sequentially reads one input symbol at a time each time applying the state-transition function to transfer itself to the next state. After all input symbols have been processed in this way, the NFA stops and outputs “true” if the last state is in the set of final states, or “false” otherwise. Notice that it is not defined how input symbols are generated, just that there is a sequence of them being fed to the automaton. Also notice that NFAs can nondeterministically transfer to a set of states at some point. This nondeterminism is important but also confusing – how can the automaton be at more than a single state? Well, although the formal definition indicates that the automaton is in a set of states, the practical and useful meaning is that the automaton may be at any single state from that set – we don’t know which one in advance, it can be any single one. In other words, a single state from the set will be chosen in some way e.g. using probabilities.

This iterative process of automaton operation can be modeled as a system:

CurrentState = s0
while there are more input symbols to process:
  InputSymbol = GetNextInput()
  CurrentState =
    TransitionFunction(InputSymbol, CurrentState)


Here, CurrentState is a component that stores the current state of the automaton, InputSymbol is a component that stores the current input symbol (element from the input alphabet), GetNextInput is a component that provides the next input symbol and writes it to the InputSymbol component and TransitionFunction is a component that computes the state transition function and writes the new state to the CurrentState component. Notice that the GetNextInput component doesn’t take any input in this description. This is not entirely true, of course – the operation of the component that generates input symbols is out of scope of formal automaton definition, the component may generate input symbols based on whatever it wants. So, I’ll expand the system to include the CurrentState as an input to the GetNextInput component (since there is nothing else in the model to include):

CurrentState = s0
while there are more input symbols to process:
  InputSymbol = GetNextInput(CurrentState)
  CurrentState =
    TransitionFunction(InputSymbol, CurrentState)

Now the interesting part – mapping properties of RESTful user-agents to this NFA model. As I wrote before, this is a simplified first-draft-quality model, not everything that should be included is included and thus the model may change when more details do get included. First I’ll try to explain the mapping in a general way, not on a specific RESTful system, and after that I’ll give a concrete example:

  • The input alphabet (set of input symbols generated by the GetNextInput component and stored in the InputSymbol component) are REST requests. Since a REST requests is defined by a resource identifier, method to be performed on the resource, metadata and a possible representation, the input alphabet is the set of all possible (syntactically) well-formed REST requests. This is why the resource identifier is not the only thing representing a transition – a single transition from a specific state is defined by all elements of a REST request, only one of which is the resource identifier. Therefore, there may be multiple transitions from a specific state and a specific destination resource identifier.
  • The state transition function, implemented in the TransitionFunction component, is the cumulative processing of REST requests. This process is performed by user-agents, intermediaries and origin servers, and the result is a REST response containing metadata and a representation of the resource identified in the related REST request. Here’s the most important part – since the user-agent doesn’t know what the response to a request will be and since the origin server may return different REST responses for the same REST request – this processing must be modeled as nondeterministic. In other words, the result of the transition function is a set of all possible REST responses which may be returned for a given REST request. As I explained earlier when demystifying nondeterminism, the origin server will in fact return a single response but since we don’t know which one – this is modeled as a set of possible responses. Also, the metadata (media type) of the response defines if the state is steady or not by specifying which of the resources linked to from the response representation should also be requested. If the state is not steady, more requests should be sent for those resources. If the input symbol is a REST request for a resource for which there is no link in the current state then the transition function returns an empty set. The same happens if the system is not in a steady state and the input symbol is a REST request not in the set of pending requests.
  • The current state, stored in the CurrentState component, is a set of REST responses and pending REST requests. The current state is considered a steady state if there are no pending requests, otherwise it’s considered a transient state. The initial state of the user-agent is steady and may contain a predefined representation with links to bootstrap the operation of the user-agent.
  • The only thing left is to define how the user-agent chooses transitions – the next REST request for the current state. This is the role of the GetNextInput component which implements both application- and hypermedia-level logic. The application-level logic is in charge of generating input symbols in case the current state is steady, or in other words, it chooses then next step for achieving the overall application goal. The application-level logic may be a software program or a program that delegates this responsibility to a human user. The hypermedia-level logic is in charge of generating input symbols in case the current state is transient, or in other words, it chooses which of the pending REST requests will be processed next. To satisfy the HATEOAS constraint, both the application- and hypermedia- level logic generate REST requests with resource identifiers linked to from the current state.

Or mathematically (without all the detailed explanations):

  • Σ = { R | R is a valid REST request consisting of a resource identifier, method, metadata and representation }
  • S = { { G } | G is a valid REST response consisting of metadata and a representation or a pending REST request }
  • s0 = initial representation containing links to bootstrap the operation of the user agent
  • δ : S x Σ → P(S) where P(S) the power set of S.
  • F = { Z | ZS and Z is a steady state }

The operation of the automaton can be described as follows. The automaton starts from the initial steady state containing links for bootstrapping the application. Each time the automaton is in a steady state, the application level logic generates the next input symbol (request) based on the links in the current state representations and the overall application goal. The request and the current state are used by the processing infrastructure (user-agent, intermediaries, origin server) to generate a response. Notice here that the current state (representation already on the client) need not be sent to the server, it may be used on the user-agent for processing (e.g. to determine if the request is for a resource linked to from the current state or not) while only the input symbol (the request) is sent over the network. The result of the processing is the new state of the user-agent. The new state may either be steady or transient, based on the representation received in the response and it’s metadata (media type). If the state is transient, the hypermedia-level logic automatically chooses requests for fetching other resources in order to bring the user-agent into a steady state so that application-level logic can take over.

Now an example – applying this model to a specific RESTful system, which will of course be a simple web application shown above. The purpose of the web application is to simulate coin tossing. It has two web pages, Main.html and Cointoss.html. The Main.html page contains only a single link to the Cointoss.html page and nothing else. The Cointoss.html page has a link back to the Main.html page and a single image. Since the web application simulates coin tossing, which is nondeterministic, the image which gets included into the Cointoss.html is chosen randomly by the server and is either the heads.png image or the tails.png image.

The input symbols of the resulting automaton are requests for fetching application resources – the html pages and the images. The states of the resulting automaton are both steady states representing a complete page load or transient states representing partial page loads. Therefore, the Main.html page is represented by only a steady state since it only contains a link to Cointoss.html, while the Cointoss.html page is represented by both a transient state and a steady state since it contains an image which must be fetched after the initial page. Let’s assume that the initial state contains a single link to the Main.html page. The transition function of the resulting automaton defines which responses the processing infrastructure returns for each state and requests. Therefore, if the user-agent requests a page or image, the server returns the requested page or image. However, if the user-agent is requesting the Cointoss.html page – it isn’t known in advance what the server will return (heads or tails), so we need a nondeterministic transition into two possible states. Lastly, steady states define the set of acceptable states.

And here’s the formal automaton definition for the explanation above:

  • Σ = { R1, R2, R3, R4 }, where R1 is a request for fetching Main.html, R2 is a request for fetching Cointoss.html, and R3 and R4 are requests for fetching the heads.png and tails.png images. For simplicity, let’s say these are the only input symbols the user-agent may generate, but in general – it may generate other syntactically valid requests.
  • S = { S1, S2, S3, S4, S5, S6}, where S1 is the initial state containing a link to Main.htmlS2 is the state after receiving the Main.html page, S3 is the state after receiving the Cointoss.html page which has a link to the heads.png image which hasn’t been fetched yet, S4 is the state after receiving responses to both the Cointoss.html page and heads.png, S5 is the state after receiving the Cointoss.html page which has a link to the tails.png image which hasn’t been fetched yet and S6 is the state after receiving responses to both the Cointoss.html page and tails.png.
  • s0 = S1, since S1 is the initial state.
  • The transition function is defined for each pair of automaton state and input symbol: δ(S1, R1) = S2δ(S2R2) = {S3, S5}, δ(S3R3) = S4δ(S4R1) = S2δ(S5R4) = S6δ(S6R1) = S2. For other pairs of state and input symbol the function returns an empty set { } since there are no other links for those pages (states) to resources identified in those requests (input symbols).
  • F = { S1, S2, S4, S6 }, since those states are steady.

And here’s the equivalent state diagram visualization of the automaton:


For such a model, we could generate a random sequence of input symbols and check if the automaton will end up in a steady state or not. You can do this for homework :).


(EDIT: inserted a few sentences to make this a real conclusion) That’s about it. All-in-all, I think REST is an important concept for the WWW and that more work should be done to explain it fully and concisely. After that, we can simplify the models to reflect the most important properties, but simplifying before understanding leads to ignorance and misunderstanding in discussions.

If you have ideas or comments about systematizing knowledge, terminology and models for REST or developing formal models for REST – let me know. @izuzak

Real-time feed processing and filtering

Posted in web development by Ivan Zuzak on January 11, 2010

This blog has moved to a new domain. You can view this post here.

This is another lengthy post so I’m writing a brief overview as an introduction.


This post is about web syndication feeds (RSS and ATOM), technologies for real-time delivery of feeds (PubSubHubBub, RSSCloud) and two opportunities I believe could help make these technologies better and widen their adoption: real-time feed processing/filtering and end-user selection of processing/filtering services. First, I’ll write about my experience with polling-based feed processing services from developing Feed-buster, a feed-enhancing service for use in FriendFeed. Next, I’ll give a short overview of PubSubHubBub (PSHB) and several ways of integrating feed filtering and processing functionalities into the PSHB ecosystem without changing the PSHB protocol. Lastly, I’ll try to argument why there should be a better solution and outline an idea for an extension to the PSHB protocol.

Before I continue, I have to acknowledge that I’ve discussed some of these ideas with Julien Genestoux (the guy behind Superfeedr) and Brett Slatkin (one of PSHB developers) several months ago. Julien and I were even starting to draft a proposal for a PSHB extension for supporting filters, however, due to other stuff in our lives, we didn’t really get anywhere.

Polling-based feed processing and filtering – experience from FriendFeed and Feed-Buster

FriendFeed is a social-networking web application that aggregates updates from social media sites and blogs that expose an RSS or Atom feed of their content. You give FriendFeed the URI of an RSS/Atom feed and it fetches and displays short snippets of updates as they appear in the feed. In most cases this means that only the title and a short snippet of the description of the new item are shown. For feeds that have special XML tags that list media resources in the update, the images/videos/audio content from the new feed item is also displayed. The problem is that most RSS/Atom feeds don’t have these media XML tags since webapps that create the feeds (e.g. blogging engines) don’t bother defining them. This gave a not-very-user-friendly look to most FriendFeed user pages which were just a long list of update titles and possibly snippets.



So I built Feed-buster, an AppEngine service that inserts these MediaRSS tags into feeds that don’t have them. Since the only thing FriendFeed accepts from users are URIs pointing to feeds – the service must be callable by requesting a single URI with a HTTP GET request and must return an RSS/Atom feed in a HTTP response. Here’s how Feed-buster works:

  1. An URI to an RSS/Atom feed is supplied as a query parameter of a HTTP GET request to the service. E.g. http://feed-buster.appspot.com/mediaInjection?inputFeedUrl=http://myfeed.com/atom.
  2. The service fetches the target feed, finds all the images, video and audio URIs and inserts the special XML tags for each one back into the feed.
  3. The service returns the modified feed as the content of a HTTP response.

To use the service, the user would obtain an URI of an RSS/Atom feed, create the corresponding Feed-buster URI by appending the feed URI to the Feed-buster service URI and pass it to FriendFeed (1). Suppose that the feed is not real-time enabled (i.e. a PSHB or RSSCloud feed). In this case, every 45 minutes FF will poll for updates – send a HTTP GET request to the service (2), the service will fetch the original feed (3), process the feed (4) and return a HTTP response with the processed feed (5).

Feed-buster architecture

Feed-buster architecture

Feed-buster is a feed processing service: it takes a feed URI as input and returns the feed it points to as output, potentially modifying any subset of feed items (but not removing any). Feed filtering services are similar: they take a feed URI as input and return the feed it points to as output, potentially removing any subset of feed items (but not modifying any). The difference is subtle and both terms could be unified under “feed processing”. Furthermore, this model of feed processing/filtering was nothing new at the time I developed Feed-buster, there were numerous services that filtered/processed feeds based on user preferences. The best example is probably Yahoo! Pipes which enables anyone to create a filtering/processing pipe a publish it as a service just like Feed-buster.

Lessons learned:

  • Feed-buster is very simple to use and this helped to make it popular (appending an URI to another URI is something everyone can do).
  • Feed-buster’s model of use is based on separation of filtering/processing services as individual entities in the ecosystem which enables that these services be developed independently of publishers and subscribers. Applications that enable other applications make ecosystems grow.
  • Polling is bad (1). Because AppEngine applications have a fixed free daily quota for consumed resources, when the number of feeds the service processed increased – the daily quota was exhausted before the end of the day because FF polls the service for each feed every 45 minutes. See the AppEngine resource graph below for a very clear visual explanation :). Of course, I implemented several caching mechanisms in the service and it helped, but not enough.
  • Polling is bad (2). Real-time feed delivery arrived in the form of PubSubHubBub and RSSCloud, everyone loved it and wanted everything to be real-time. But Feed-buster feeds weren’t real-time. Even if the original feed supported PSHB, the output feed didn’t (since it Feed-buster wasn’t a PSHB publisher or hub). Feed-buster users were not happy.
Feed-buster - AppEngine request processing duration graph

Feed-buster - AppEngine request processing duration graph


PubSubHubBub – real-time feed delivery

PubSubHubBub is a server-to-server publish-subscribe protocol for federated delivery of feed updates. Without going into a lot of detail, the basic architecture consists of publishers, subscribers and hubs which interact in the following way:

  1. A feed publisher declares its hub in its Atom or RSS XML file and a subscriber initially fetches the feed to discover the hub.
  2. The subscriber subscribes to the discovered hub declaring the feed URI it wishes to subscribe to and an URI at which it wants to receive updates.
  3. When the publisher next updates the feed, it pings the hub saying that there’s an update and the hub fetches the published feed.
  4. The hub multicasts the new/changed content out to all registered subscriber.
PubSubHubBub ecosystem and protocol

PubSubHubBub ecosystem and protocol

And this gives us real-time feed delivery. But what about real-time filtering and processing? How would a real-time-enabled version of Feed-buster be built? Of course, there are a lot more reasons for doing this, both from business and engineering perspectives.

Real-time feed processing and filtering (take 1)

So, the idea is to extend the PSHB ecosystem with filtering/processing services in some way. First, here are a few ideas for doing this without extending the PSHB protocol:

  1. Integrating filtering/processing services into publishers. This means that the publisher would expose not only the original feed, but also a feed for each filter/processing defined by the subscriber. For example, Twitter has a single feed of all updates for a specific user, a feed for updates from that user containing the string “hello”, updates not containing “world”, updates written in English, and so on.
  2. Integrating filtering/processing services into subscribers. This means that the subscriber application would subscribe to feeds and filter/process them on it’s own. It’s not exactly the same thing but, for example, FriendFeed subscribes to feeds you want it to aggregate and later you can filter the content from within the FriendFeed application — this is called search, as in Twitter, but basically means filter.
  3. Integrating filtering/processing services into hubs. This means that a hub would not only deliver feed updates, but process and filter them also. This would cause a large number of hubs to be developed, each implementing some filter or processing functionality. If you want to filter a feed – subscribe to a hub that implements that filter. If you want a new filter – build a new hub. However, if an end-user tells a subscriber application to subscribe to a feed with a specific URI, the subscriber will subscribe to a hub specified in the feed, not to the filtering/processing hub (since that hub wasn’t  specified in any way). A possible way for enabling end-users to select which hub they want to use is to generate dummy feeds on filtering/processing hubs (e.g. http://www.hubforfiltering.com/http://www.myfeed.com/atom). The contents of this dummy feed is the same as the contents of the original feed but with a simple difference – the hub elements from the original feed are removed and a hub element for new hub is inserted. This way, when the end-user passes the dummy feed URI to the subscriber application, the application will subscribe to the filtering/processing hub, not the original one. The filtering/processing hub then subscribes to the original hub to receive updates, filters/processes received updates and notifies the subscriber.
  4. Masking the filtering/processing services as the subscriber which then notifies the real subscriber. I’m not sure this is possible (too lazy to go through the PSHB spec), but in theory, here’s how it would work. The subscriber would obtain an URI pointing to the filtering/processing service for which the service would know what the real subscriber notification callback is (e.g. http://www.filter.com?subscriber=http://www.realsubscriber.com/callback). The subscriber would subscribe to the hub using the obtained URI. When the publisher publishes an update, the hub would pass the update to the filtering/processing service thinking it was the subscriber. The filtering/processing service would do it’s thing and pass the update to the real subscriber. It’s unclear though how this would be done by end-users since they’d have to know the subscriber application’s callback URI.

All of these approaches have their pros and cons. Some are not easily or at all end-user definable, some may introduce unnecessary duplication of notifications, some introduce delivery of unwanted notifications, some do not support reuse of existing infrastructure and cause unnecessary duplication of systems, and most do not promote the development of filtering/processing services as a separate part of the ecosystem.

Real-time feed processing and filtering (take 2)

What I believe is needed is an architecture and protocol which supports three things. First – filtering/processing services as a separate part of the ecosystem. Since publishers and subscribers can’t possibly anticipate and implement every possible filter and processing service, this will enable a stream of development activity resulting in services with different functionality, performance and business models. Again, applications enabling other applications help software ecosystems grow.

Second – a scalable processing model where hubs invoke filtering/processing services. Regarding architectural decisions on where filtering/processing should be applied, there’s a paper (citation, pdf) from 2000. on “Achieving Scalability and Expressiveness in an Internet-scale Event Notification Service” which states two principles:

Downstream replication: A notification should be routed in one copy as far as possible and should be replicated only downstream, that is, as close as possible to the parties that are interested in it.

Upstream evaluation: Filters are applied and patterns are assembled upstream, that is, as close as possible to the sources of (patterns of) notifications.

In short, these principles state that filtering services should be applied as close to the publisher as possible so notifications that nobody wants don’t waste network resources. However, processing services should be applied as close to the subscriber so that the original update may be transported through the network as a single notification for as long as possible. In cases where there’s a single hub between the publisher and subscriber, the same hub would invoke both processing and filtering services. However, in more complex hub network topologies different hubs would invoke different types of services based on their topological proximity to the subscriber/publisher.

Third and last – end-user selection of these services. When 3rd party filtering/processing services exist – how will they be chosen for a particular subscription? In my opinion, this must be done by end-users and thus there must be an easy way of doing it. The first thing that comes to mind would be to extend the subscription user-interface (UI) exposed to users of subscriber applications. E.g. instead of just showing a textbox for defining the feed the users wants to follow, there could be an “Advanced” menu with textboxes for defining filtering and processing services.

New PSHB architecture and protocol outline

New PSHB architecture and protocol outline

No matter how cool I think this would be as a core part of the PSHB protocol, it’s probably a better idea doing it as an extension (as a first step?). So here’s what the extension would define:

  1. The subscription part of the protocol additionally enables subscribers to define two lists of URIs when subscribing to a hub, one declaring the list of filtering services and the other the list of processing services that should be applied.
  2. Hubs are responsible not only for delivery of notifications but also for calling filtering/processing services on those notifications and receiving their responses. An inter-hub subscription negotiation/propagation mechanism that enables the optimal placement of subscription-defined filters and processing services in the delivery flow. In case of a single hub, nothing needs to be done since the hub invokes both types of services. However, in case of two hubs, the hub which first received the subscription would locally subscribe to the second hub stating only the filtering services while processing services would not be sent upstream. Thus, the hub closer to the subscriber would be responsible for invoking processing services, and the hub closer to the publisher would be responsible for invoking filtering services. This could, I believe, be generalized and formalized for any kind of hub network topology.
  3. When an update is published, hubs responsible for invoking filtering or processing services must do so before forwarding the notifications downstream. There are several performance-related issues which need to be addressed here. First, the invocation of filtering/processing services should be asynchronous, webhook-based (the hub calls a service passing an URI to which the result of processing should be POSTed). Second, invocation of services on a specific hub for a specific notification should be done in parallel, except if multiple processing services were defined for a single subscription in which case those processing services must be invoked serially (filters can always be invoked in parallel). Third, there’s a not-so-clear trade-off between a) waiting for all invoked filters to respond and sending a single notification carrying all metadata regarding which filters it satisfied and b) sending  multiple notifications containing the same content but different metadata regarding which filters were satisfied (e.g. as each filter responds). Since all filters will not respond at the same time (some will respond sooner, some later and some possibly never) – waiting for all filters invoked in parallel before continuing seems contradictory to the constraint of real-time delivery. On the other hand, creating multiple notifications with the same data will introduce an overhead in the network. Overall, I believe that the overhead will not be substantial since there will not be many duplicates — this can even be regulated as a hub-level parameter e.g. one notification for services returning in under 2 seconds, one for under 10 second, and one for everything else, and the benefit is basically a necessity.


Of course I didn’t consider all aspects of the problem and there are a lot more points of view to be considered, from is it legal to change content in a feed and in what way to how do we make this extension secure and fault tolerant? Nevertheless, I believe that the need and expertise are here and that it’s an opportunity to make the real time web better. What are your ideas for extending real-time content delivery technologies? @izuzak

RPC for Web Workers and distributed computing within the browser

Posted in web development by Ivan Zuzak on December 21, 2009

This blog has moved to a new domain. You can view this post here.

In a previous post I wrote about pmrpc – a project of mine for developing a JavaScript library for cross-domain inter-window RPC-style communication in HTML5 browsers. RPC communication is based on two technologies: HTML5 cross-domain messaging as the transport mechanism and JSON-RPC as the RPC protocol and message format, while the library supports other advanced features (access control, asynchronous calls, retries…). At the end of the post I outlined a few things Marko and I wanted to implement next in pmrpc, one of which was support for Web Workers. From the post:

Web Workers are another specification being developed in parallel with HTML5. The specification “defines an API for running scripts in the background independently of any user interface scripts”. What’s interesting is that the postMessage API is used to communicate with web workers and therefor pmrpc could be extended to include support for web workers.

Well, the API used to send messages from a web page to a worker and vice versa is not exactly the same as the cross-document messaging API, but is very similar. So, the idea for supporting Web Workers was leave the pmrpc API for registering, calling and unregistering procedure unchanged, and just extend it to also use the Web Worker API for message transport (in addition to the cross-doc messaging). And about two weeks ago we released a new version of pmrpc with Web Worker support. Here’s an example of using pmrpc with Web Workers:

// [worker object A - testingWorker.js]

// load pmrpc library

// expose a procedure
pmrpc.register( {
publicProcedureName : "HelloPMRPC",
procedure : function(printParam) {
              alert(printParam); } } );

// [window object B]

// load pmrpc library

// create worker
var testingWorker =
  new Worker("testingWorker.js");

// calls the exposed procedure
pmrpc.call( {
destination : testingWorker,
publicProcedureName : "HelloPMRPC",
params : ["Hello World!"] } );

A few notes on worker support. Pmrpc currently supports only normal (“dedicated”) Web Workers. Shared Workers (workers shared between multiple web pages) and nested workers (workers within workers) are currently not supported since, as far as I know, Shared and nested Workers are not implemented yet in any browser. Also, since Workers must be loaded from the same origin as the page that created them – the access control feature in pmrpc make no sense and are therefore not supported for Web Workers.

Next on our list is enabling discovery of registered procedures scattered over multiple browser workers, iframes and windows. Currently, pmrpc requires that you tell it where the procedure you are calling is, but sometimes you don’t know where it is, or don’t care – e.g. you just want to call the “DrawMap” procedure loaded from domain “http://www.good.com” if it exists. Discovery should enable discovery of methods based just on their names. The problems in implementing this feature (knowing which windows, iframes and workers are currently running and syncing the list of registered procedures with each of them) made me think about where systems running in browsers are heading – towards a form of distributed & parallel computing, only within a single browser. Distributed and concurrent execution, discovery, access control, synchronization, remote invocation, fault tolerance and naming are all concepts well know in distributed computing, and will possibly be re-invented in the context of web browsers to support this new shift. I’m not sure how far pmrpc will go in this direction, but I hope it will be a good example for other developers and research projects.

Do you think that web browsers need more powerful mechanisms supporting this new form of distributed computing? @izuzak