Two-Way Web: Can You Stream In Both Directions?

Update (couple of hrs later): I mailed Alex Russell (the guy who named Comet and knows plenty about it), it sounds like he’s been investigating this whole area and he’s sent me his views.

We know about Comet (AKA Push, HTTP Streaming) and its ability to keep streaming info from server to browser. How about streaming upwards, from browser to server, and preferably in the same connection? A reader mailed me this query:

Im missing one demo, would it be possible to reuse same stream in streaming demos to send msg to server, I’ve been digging throw your examples, but they all seem to create a new connection to the server when posting, would be very interesting see a demo that does this within the same stream, and of course the server code would be as interesting as the client.

Here’s my thinking, I’m sure a lot of smart readers will know more about this and I’ll be interested in your views – is it feasible? Any online demos?

Unfortunately, I’ve not seen anyone pull this off – it’s always assumed you need a “back channel”. It’s the kind of hack someone like Google or 37S would turn around and pull off even though it’s “obviously impossible” 😉 .

There are two key issues:

(1) Server needs to start outputting before incoming request is finished. With a specialised server, this problem could be overcome.

(2) (More serious as we can’t control the browser) The browser would need to upload data in a continuous stream. You can do it with Flash/Java, but I can’t see how to do this with standard JS/HTML. If you use XHR, you’re going to call send() and wave goodbye to the entire request…there’s no support for sequencing it. Same if you submit a regular form, change IFrame’s source etc. Even if you could somehow delay reading of content so it’s not immediately uploaded, the browser would probably end up not sending anything at all as it would be waiting to fill up a packet.

I think the solution lies in the Keep-Alive extension to HTTP 1.1:

What is Keep-Alive?

The Keep-Alive extension to HTTP, as defined by the HTTP/1.1 draft, allows persistent connections. These long-lived HTTP sessions allow multiple requests to be send over the same TCP connection, and in some cases have been shown to result in an almost 50% speedup in latency times for HTML documents with lots of images.

If you google for “xmlhttprequest keep-alive” or “ajax keep-alive”, you’ll see people talking about the idea a bit, but there’s not much info on how to script it for continuous connections and no demos to be found. It would make a great experiment if someone did a proof-of-concept!

As an alternative, you could consider a thin, invisible, Flash layer to handle transport, and degrade to frequent Submission Throttling where Flash isn’t an option.


p>BTW I have a post and podcast planned about the whole two-way web thing, which will be profound (the two-way web thing, not the podcast :-)). The web is entering a new era of Real-Time Collaboration and Communication, post-Ajax (and of course building on Ajax, just as Ajax builds on the technologies of the previous era: CGI, DHTML, CSS, etc).

Update: As mentioned above, Alex Russell mailed me his views. In particular, it’s interesting to consider the possibility that browsers might transparently exploit keep-alive if you hit the server frequently enough.

So I’ve spent some time investigating this (as you might expect), and at the end of the day there’s not much to be done aside from using Flash and their XMLSocket interface. That’s an obvious possibility given the high-performance Flash communication infrastructure we have in Dojo. Doing bi-directional HTTP probably won’t happen, though, but I don’t think that’s cause for despair. In my tests, we can get really good (relative) performance out of distinct HTTP requests so long as the content of the request is kept to a minimum and the server can process the connection fast enough. HTTP keepalive exists at a level somewhat below what’s currently exposed to browsers, so if the client and server support it, frequent requests through stock XHR objects may verywell be using it anyway. We’ll have to do some significant testing to determine what conjunctions of servers/clients might do this, however.

There are even more exotic approaches available from Flash peering that I’ve been investigating as well, but they will require significantly different infrastructure from what we already deploy that I think they’re still in the land of “hrm…someday”.

First we have to solve the *regular* Comet scalability problems for existing servers and app containers.


PS: we haven’t been making much noise about it, but serious work has started on an Open Source Comet protocol with initial implmemntations in both Perl and Python over at The initial client library is Dojo-based, but we’ll be publishing the protocol so that anyone can “play” with it.

Portable Comet? It’s the IFrame, Stupid!

Comet Takes to IE Like a Fish Takes to Acid

Comet – or HTTP Streaming, if you will – is a little sensitive when it comes to portability, and I’ll give you four guesses which major browser is causing the grief? Yeah, IE makes it difficult for two reasons: (a) IE’s XMLHttpRequest component doesn’t tell you anything about the response until the connection has closed – even if you try polling it instead of relying on onReadyStateChange, you’ll still get an empty string (Try it); (B) Okay, switch to plan B and inspect IFrame content – we can’t rely on onload, which is only called once at the end, so we must poll. But no, polling won’t help either, as the IFrame content remains empty until (you guessed it) the connection is closed. (Try it).

We’re Going Back, Way Back: Inline Script Tags

Don’t give up on the IFrame so fast … we’re closer than we think. Actually, the solution harkens back to one of the original Push techniques: outputting script tags with inline code (what the HTTP Streaming pattern calls “page streaming”). If you do that in your main document, the code will be executed immediately, and the same is true if you do that inside an IFrame.

I’m talking about a service that outputs stuff like this:

  1. <script type="text/javascript">
  2.     doSomething();
  3. </script>

So Make the Server Spit Out Inline Script Tags

The portable solution is this: Have the server continuously output script tags that call a known function in the parent frame. When you set the child IFrame’s source to point to this service, it will start evaluating the inline scripts as they pop out of the server. This happens to be one technique people have used for remoting for many years (I think Brent Ashley recently told me he was doing it in ?1999). The twist with Comet is that you keep doing it, and don’t actually close the connection for a while. (Again, I’m sure some people were doing that for a long time too!).

Is it elegant? No. It means the remote service is suddenly coupled to the client-side of the web app – it has to know something about what’s in the parent frame, whereas you’d often like it to be generic and just deliver a response in XML, JSON or whatever. Like most things Ajax, we’re using a hack because it nonetheless push all the right buttons for our users.

Try It. Portable Comet for the Masses

Here’s Some Code I Prepared Earlier

In the example above, the remote service outputs script tags in a loop – it’s a simple countdown service:

  1. <?
  2.   for ($i=$_GET&#91;"start"]; $i>=0; $i--) {
  3. ?>
  4.   <script type="text/javascript">
  5.     window.parent.onNewValue(<?= $i ?>);
  6.   </script>
  7. <?
  8.   }
  9. ?>

And in the browser’s initial Javascript, there’s a run-of-the-mill onNewValue() function, it looks something like this.


  1. function onNewValue(i) {
  2.   statusDiv.innerHTML = i; // Paint the new value onto the page
  3. }

See what I mean about coupling? The server-side service had to know that there’s a Javascript function defined in the parent called onNewValue(). At least we’ve minimised the coupling by using an Observer/Event style indirection approach – evidenced by the simple call to “onNewValue()”. It would be worse if it was the script that actually performed application logic directly, repainting the DOM with the new value.

IFrame is the new XMLHttpRequest

Whoever put the X in Ajax ought to put an I in Comit. IE’s XHR component doesn’t provide the response until the connection has closed, and I don’t think version 7 changes that. Assuming (graciously) it happens in IE8, you’ll need to use IFrame for the next five years if you’re doing any Comet work. And of course, you won’t need to do that, because libraries will do it for you.

BTW You could argue that IE is doing the right thing by preventing access to the content until the connection is closed. Maybe they are, maybe they aren’t. From a pragmatic perspective, though, portable Comet requires Ajax. Alternatively, use IFrame with IE and XmlHttpRequest with the others, though I’m not sure if there’s much mileage to be gained from this hybrid strategy.

Comet: It’s Ajax for “Push” (Podcast)

Here’s a podcast about Comet – exploring the two-way web with Ajax. From my Ajaxian post earlier today:

Alex Russell has coined a term for a flavour of Ajax that’s been getting more attention of late. Comet describes applications where the server keeps pushing – or streaming – data to the client, instead of having the browser keep polling the server for fresh content. Alex identifies several buzzworthy examples:

This is an important article because it captures a growing trend in Ajax, a trend I had in mind when I said we expect to hear more about “Push and the Two-Way Web” in the next twelve months, on the occasion of Ajax’s birthday. There will, of course, be people saying “there’s nothing new here”, and that’s presumably all too obvious to Alex himself, who has worked with these ideas for a long time. But as with Ajax, it’s the power of a name. I don’t think these ideas can adequately be described as Ajax, because Ajax changes a lot about the browser whereas Comet fundamentally changes the nature of browser-server communication. I see Comet as part of the overall Ajax trend, complementary to the UI aspects of Ajax.

People may also say there are existing names like “Push”. True, but they have baggage – I think it’s useful to have a name for this architectural pattern in light of the relationship to Ajax.

Anyways, I wanted to expand on some of the thoughts in the article and after the recent Basics of Ajax Podcast, I’m in the mood for more audio rambling. So here’s a 56-minute discussion about Comet and the general trend of push and streaming within Ajax.

Click to download the Podcast. You can also subscribe to the
feed if you want future podcasts automatically downloaded - check out the
podcast FAQ at


It's the Duplex, Stupid! Push or Pull - it doesn't matter so much. What's critical here is the focus on the two-way web.

Applications - Chat - Wiki - News - Current events, sport, financials, etc - Trading and Auctions - Real-time control and logistics - File transfer (combine with local storage) - Any other genre you'd care to name

Vanilla Ajax: Await the User

Comet Ajax: Keep Pushing

Polling Ajax: Keep Pulling

Benefits of Comet - Responsive: data pumped out immediately - More stable profile - Less overhead of establishing connections

Benefits of Polling - Browser memory - Can run on any standard server; Comet requires suitable server - Can upload at the same time - Can run on - with Comet, XHR and IFrame won't always reflect changes while the connection's open - Being more standard, works with existing infrastructure. Comet is vulnerable to middle-men dropping the connection. - Simpler architecture - only the browser's in control - Easier to test - More familiar architecture - Less programming effort - with Comet, must watch for changes on the stream - More efficient for infrequently accessed data - Leverages caching

Maybe Comet causes more pain, but if it keeps the user happy ...

Questions and Trends - Which to use. Variables include: frequency of updates, importance of updates, server capabilities, target browsers - Dealing with incoming messages, e.g. Distributed Events pattern, Event bus (browser or server?), etc - Workarounds for throbber, status bar, clicking sound, etc. - How often to drop connections - How browsers can accommodate it

Proof-Of-Concept Demos - Wiki using Periodic Refresh/Polling - Wiki using HTTP Streauh, Comet (Actually, this is only a very basi implementation - there's no use of events, just custom handling of HTTP.

Related Patterns - HTTP Streaming - Periodic Refresh (aka Polling) - Distributed Events

As always, feedback is welcome – [email protected]