Introduction
Microsoft recently released Beta 2 of ASP.NET AJAX. Although it's a very powerful framework, when you will build a real AJAX site like those out there in the Web 2.0 world, you will face many problems that you will hardly find documented anywhere. In this article, I will show some advance level ideas that I learned while building Pageflakes. We will look at the advantages and disadvantages of Batch calls, AJAX call timeouts, browser call jam problem, ASP.NET 2.0's bug in web service response caching, and so on.
Why use ASP.NET AJAX
When others see Pageflakes, the first question they ask me is: "Why did not you use Protopage or Dojo library? Why Atlas?" Microsoft Atlas (renamed to ASP.NET AJAX) is a very promising AJAX framework. They are putting a lot of effort on it, making lots of reusable components that can really save you a lot of time and give your web application a complete face lift at reasonably low effort or changes. It integrates with ASP.NET very well, and it is compatible with the ASP.NET Membership and Profile provider. The AJAX Control Toolkit project contains 28 extenders which you can drag & drop on your page, tweak some properties, and add pretty cool effects on the page. Check out the examples to see how powerful the ASP.NET AJAX framework has really become.
When we first started developing Pageflakes, Atlas was in infant stage. We were only able to use the page method and Web Service method call features of Atlas. We had to make our own drag & drop, component architecture, popups, collapse/expand features etc. But now, you can have all these from Atlas and thus save a lot of development time. The web service proxy feature of Atlas is a marvel. You can point a <script> tag to a .asmx file and you get a JavaScript class generated right out of the web service definition. The JavaScript class contains the exact methods that you have on the web service class. This makes it really easy to add/remove new web services, and add/remove methods in web services which does not require any changes on the client side. It also offers a lot of control over the AJAX calls, and provides rich exception trapping feature on the JavaScript. Server side exceptions are nicely thrown to the client side JavaScript code, and you can trap them and show nicely formatted error messages to the user. Atlas works really well with ASP.NET 2.0, eliminating the integration problem completely. You need not worry about authentication and authorization on page methods and web service methods. So, you save a lot of code on the client side (of course, the Atlas Runtime is huge for this reason), and you can concentrate more on your own code than building up all these framework related code.
The recent version of Atlas works nicely with ASP.NET Membership and Profile services, giving you login/logout features from JavaScript without requiring page postbacks, and you can read/write Profile objects directly from JavaScript. This comes very handy when you heavily use ASP.NET membership and profile providers in your web application, which we do at Pageflakes.
On earlier versions of Atlas, there was no way to make HTTP GET calls. All calls were HTTP POST, and thus quite expensive calls. Now, you can say which calls should be HTTP GET. Once you have HTTP GET, you can utilize HTTP response caching features which I will show you soon.
Batch calls are not always faster
ASP.NET AJAX has a feature in the CTP release (and previous releases) which allows batching multiple requests into one request. It works transparently, you won't notice anything, nor would you need to write any special code. Once you turn on the Batch feature, all web service calls made within a duration gets batched into one call. Thus it saves roundtrip time and total response time.
The actual response time might be reduced, but the perceived delay is higher. If three web service calls are batched, the first call does not finish first. All three calls finish at the same time. If you are doing some UI updates upon completion of each WS call, it does not happen one by one. All of the calls complete in one shot and then the UI gets updated in one shot. As a result, you do not see incremental updates on the UI, instead a long delay before the UI updates. If any of the calls, say the third call downloads a lot of data, the user sees nothing happening until all three calls complete. So, the duration of the first call becomes nearly the duration of the sum of all three calls. Although the actual total duration is reduced, the perceived duration is higher. Batch calls are handy when each call is transmitting a small amount of data. Thus three small calls get executed in one roundtrip.
Let's work on a scenario where three calls are made one by one. Here's how the calls actually get executed.
The second call takes a little bit time to reach the server because the first call is eating up the bandwidth. For the same reason, it takes longer to download. Browsers open two simultaneous connections to the server. So at a time, only two calls are made. Once the second/first call completes, the third call is made.
When these three calls are batched into one:
Here the total download time is reduced (if IIS compression is enabled), and there's only one network latency overhead. All three calls get executed on the server in one shot, and the combined response is downloaded in one call. But to the user, the perceived speed is slower because all the UI update happens after the entire batch call completes. The total duration the batch call will take to complete will always be higher than that for two calls. Moreover, if you do a lot of UI updates one after another, Internet Explorer freezes for a while, giving the user a bad impression. Sometimes, expensive updates on the UI makes the browser screen go blank and white. But Firefox and Opera does not have this problem.
Batch calls have some advantages too. The total download time is less than that for downloading individual call responses because if you use gzip compression in IIS, the total result is compressed instead of individually compressing each result. So, generally, a batch call is better for small calls. But if a call is going to send a large amount of data or is going to return, say, 20KB of response, then it's better not to use batch. Another problem with batch call is, say two calls are very small but the third call is quite big, and if these three calls get batched, the smaller calls are going to suffer from the long delay due to the third larger call.
Bad calls make good calls timeout
If two HTTP calls somehow get stuck for too long, those two bad calls are going to make some good calls expire too, which in the meantime got queued. Here's a nice example:
function TestTimeout()
{
TestService.set_defaultFailedCallback( function(result,
userContext, methodName)
{
var timedOut = result.get_timedOut();
if( timedOut )
debug.trace( "Timedout: " + methodName );
else
debug.trace( "Error: " + methodName );
});
TestService.set_defaultSucceededCallback( function(result)
{
debug.trace( result );
});
TestService.set_timeout(5000);
TestService.HelloWorld("Call 1");
TestService.Timeout("Call 2");
TestService.Timeout("Call 3");
TestService.HelloWorld("Call 4");
TestService.HelloWorld("Call 5");
TestService.HelloWorld("Call 6");
}
On the server side, the web service is very simple:
[WebService(Namespace = "http://tempuri.org/")]
[WebServiceBinding(ConformsTo = WsiProfiles.BasicProfile1_1)]
[ScriptService]
public class TestService : System.Web.Services.WebService {
public TestService () {
//Uncomment the following line if using designed components
//InitializeComponent();
}
[WebMethod][ScriptMethod(UseHttpGet=true)]
public string HelloWorld(string param) {
Thread.Sleep(1000);
return param;
}
[WebMethod][ScriptMethod(UseHttpGet=true)]
public string Timeout(string param) {
Thread.Sleep(10000);
return param;
}
}
I am calling a method named "Timeout" on the server which does nothing but wait for a long time so that the call gets timed out. After that, I am calling a method which does not timeout. But guess what the output is:
Only the first call succeeded. So, if at any moment, the browser's two connections get jammed, then you can expect other waiting calls are going to timeout as well.
In Pageflakes, we used to get nearly 400 to 600 timeout error reports from users' browsers. We could never find out how this could happen. First, we suspected slow internet connection. But that cannot happen for so many users. Then we suspected something is wrong with the hosting provider's network. We did a lot of network analysis to find out whether there were any problems on the network or not. But we could not detect any. We used SQL Profiler to see whether there were any long running queries which timed out the ASP.NET request execution time. But no luck. We finally discovered that, it mostly happened due to some bad calls which got stuck and made the good calls expire too. So, we modified the Atlas Runtime and introduced automatic retry on it, and the problem disappeared completely. However, this auto retry requires a sophisticated open heart bypass surgery on the ASP.NET AJAX framework JavaScript. The idea is to make each and every call retry once when it times out. In order to do that, we need to intercept all web method calls and implement a hook on the onFailed callback, which will call the same web method again if the failure reason was a timeout.
Another interesting discovery we made while we were traveling was that whenever we tried to visit Pageflakes from a hotel or an airport wireless internet connection, the first visit always failed and all the web service calls on first attempt always failed. Until we did a refresh, nothing worked. This was another major reason why we implemented immediate auto retry of web service calls, which fixed the problem.
Here's how to do it. The WebMethod class in the ASP.NET AJAX framework has an internal method named invokeInternal which is called when a web method needs to make a web request. So, we replace this function with a custom implementation which passes a custom onFailure callback. That custom callback gets fired whenever there's an error or timeout. So, when there's a time out, it calls the web method again.
CollapseSys.Net._WebMethod.prototype._retryOnFailure =
function(result, userContext, methodName)
{
if( result.get_timedOut() )
{
if( this.isRetry )
{
debug.trace("Retry: " + methodName);
this.isRetry = false;
this._originalInvokeInternal(this.params,
this.onSuccess, this.onFailure,
this.userContext );
}
else
{
onFailure(result, userContext, methodName);
}
}
else
{
onFailure(result, userContext, methodName);
}
}
Sys.Net._WebMethod.prototype._originalInvokeInternal =
Sys.Net._WebMethod.prototype._invokeInternal;
Sys.Net._WebMethod.prototype._invokeInternal =
function(params, onSuccess, onFailure, userContext)
{
this.onSuccess = onSuccess;
this.onFailure = onFailure;
this.userContext = userContext;
this.params = params;
this.isRetry = true;
Sys.Net._WebMethod.prototype._originalInvokeInternal.apply(this,
[params, onSuccess, Function.createDelegate( this,
this._retryOnFailure ), userContext]);
}
When run, it will retry each timed out call once.
Here you see the first method succeeded and all the others timed out and was retried. The reason why they all timed out again is because the methods are written to time out always. So, the situation did not change. But in a real world application, you can expect that the retry will have more probability of succeeding.
Browsers allow two calls at a time and don't expect any order
Browsers make two concurrent AJAX calls at a time to a domain. If you make five AJAX calls, the browser is going to make two calls first, then wait for any one of them to complete and then make another call until all remaining four calls are complete. Moreover, you cannot expect calls to execute in the same order as you make the calls. Here's why:
Here you see, call 3's response download is quite big, and thus takes longer than call 5. So, call 5 actually gets executed before call 3.
So, the world of HTTP is unpredictable.
Browsers do not respond when more than two calls are in queue
Try this, go to any start page in the world which will load a lot of RSS on the first visit (e.g., Pageflakes, Netvibes, Protopage), and while loading, try to click on a link which will take you to another site or try to visit another site. You will see the browser is stuck. Until all queued AJAX calls in the browser completes, the browser will not accept any other activity. This is worst in Internet Explorer. But Firefox and Opera do not have this much of a problem.
The problem is, when you make a lot of AJAX calls, the browser keeps all calls in a queue and executes two at a time. So, if you click on something or try to navigate to another site, the browser has to wait for running calls to complete before it can take another call. The solution to this problem is to prevent more than two calls being queued in browser at a time. We need to maintain a queue ourselves, and send calls to the browser's queue from our queue one by one.
The solution is quite shocking, brace for impact:
Collapsevar GlobalCallQueue = {
// Maintains the list of webmethods to call
_callQueue : [],
// Number of calls currently in progress by browser
_callInProgress : 0,
// Max number of calls to execute at a time
_maxConcurrentCall : 2,
// Delay between execution of calls
_delayBetweenCalls : 50,
call : function(webMethod, params, onSuccess,
onFailure, userContext)
{
var queuedCall = new QueuedCall(webMethod, params,
onSuccess, onFailure, userContext);
Array.add(GlobalCallQueue._callQueue,queuedCall);
GlobalCallQueue.run();
},
run : function()
{
/// Execute a call from the call queue
if( 0 == GlobalCallQueue._callQueue.length ) return;
if( GlobalCallQueue._callInProgress <
GlobalCallQueue._maxConcurrentCall )
{
GlobalCallQueue._callInProgress ++;
// Get the first call queued
var queuedCall = GlobalCallQueue._callQueue[0];
Array.removeAt( GlobalCallQueue._callQueue, 0 );
// Call the web method
queuedCall.execute();
}
else
{
// cannot run another call. Maximum concurrent
// webservice method call in progress
}
},
callComplete : function()
{
GlobalCallQueue._callInProgress --;
GlobalCallQueue.run();
}
};
QueuedCall = function( webMethod, params, onSuccess,
onFailure, userContext )
{
this._webMethod = webMethod;
this._params = params;
if (onSuccess === null || typeof onSuccess === 'undefined')
onSuccess =
this._webMethod._proxy.get_defaultSucceededCallback();
if (onFailure === null || typeof onFailure === 'undefined')
onFailure =
this._webMethod._proxy.get_defaultFailedCallback();
if (userContext === null || typeof userContext === 'undefined')
userContext =
this._webMethod._proxy.get_defaultUserContext();
this._onSuccess = onSuccess;
this._onFailure = onFailure;
this._userContext = userContext;
}
QueuedCall.prototype =
{
execute : function()
{
/// Execute the actual web method
this._webMethod._originalInvokeInternal.apply( this._webMethod,
[this._params,
// Handle call complete
Function.createDelegate(this, this.onSuccess),
// Handle call complete
Function.createDelegate(this, this.onFailure),
this._userContext] );
},
onSuccess : function()
{
this._onSuccess.apply( this._webMethod, arguments );
GlobalCallQueue.callComplete();
},
onFailure : function()
{
this._onFailure.apply( this._webMethod, arguments );
GlobalCallQueue.callComplete();
}
};
QueuedCall encapsulates one web method call. It takes a Sys.Net._WebMethod class instance and overrides the onSuccess and onFailure callbacks. We want to know when a call completes or fails so that we can issue another call from our queue. The GlobalCallQueue maintains the list of all calls. Whenever a web method is called, we first queue the call in the GlobalCallQueue and executes calls from the queue one by one.
In order to enable the queue based call, we need to override the ASP.NET AJAX web method invocation again, as we did before.
Sys.Net._WebMethod.prototype._originalInvokeInternal =
Sys.Net._WebMethod.prototype._invokeInternal;
Sys.Net._WebMethod.prototype._invokeInternal =
function(params, onSuccess, onFailure, userContext)
{
GlobalCallQueue.call(this, params, onSuccess, onFailure, userContext);
}
Caching web service response on the browser and saving bandwidth significantly
Browsers can cache images, JavaScript, CSS files on a user's hard drive, and it can also cache XML HTTP calls if the call is a HTTP GET. The cache is based on the URL. If it's the same URL, and it's cached on the computer, then the response is loaded from the cache, not from the server when it is requested again. Basically, the browser can cache any HTTP GET call and return cached data based on the URL. If you make an XML HTTP call as HTTP GET and the server returns some special header which informs the browser to cache the response, on future calls, the response will be immediately returned from the cache and thus saves the delay of network roundtrip and download time.
At Pageflakes, we cache the user's state so that when the user visits again the following day, the user gets a cached page which loads instantly from the browser cache, not from the server. Thus the second time load becomes very fast. We also cache several small parts of the page which appears on user's actions. When the user does the same action again, a cached result is loaded immediately from the local cache and thus saves the network roundtrip time. The user gets a fast loading site and a very responsive site. The perceived speed increases dramatically.
The idea is to make HTTP GET calls while making Atlas web service calls and return some specific HTTP Response headers which tell the browser to cache the response for some specific duration. If you return the "Expires" header during the response, the browser will cache the XML HTTP response. There are two headers that you need to return with the response which will instruct the browser to cache the response:
HTTP/1.1 200 OK
Expires: Fri, 1 Jan 2030
Cache-Control: public
This will instruct the browser to cache the response till Jan 2030. As long as you make the same XML HTTP call with the same parameters, you will get cached response from the computer and no call will go to the server. There are more advanced ways to get further control over response caching. For example, here is a header which will instruct the browser to cache for 60 seconds but not contact the server and get a fresh response after 60 seconds. It will also prevent proxies from returning cached response when the browser local cache expires after 60 seconds.
HTTP/1.1 200 OK
Cache-Control: private, must-revalidate, proxy-revalidate, max-age=60
Let's try to produce such response headers from an ASP.NET web service call:
[WebMethod][ScriptMethod(UseHttpGet=true)]
public string CachedGet()
{
TimeSpan cacheDuration = TimeSpan.FromMinutes(1);
Context.Response.Cache.SetCacheability(HttpCacheability.Public);
Context.Response.Cache.SetExpires(DateTime.Now.Add(cacheDuration));
Context.Response.Cache.SetMaxAge(cacheDuration);
Context.Response.Cache.AppendCacheExtension(
"must-revalidate, proxy-revalidate");
return DateTime.Now.ToString();
}
This will result in the following response headers:
The Expires header is set properly. But the problem is with the Cache-control. It is showing that "max-age" is set to zero which will prevent the browser from doing any kind of caching. If you seriously want to prevent caching, you should emit such a cache-control header. Looks like exactly the opposite thing happened.
The output is as usual incorrect, and not cached:
There's a bug in ASP.NET 2.0 that you cannot change the "max-age" header. As max-age is set to zero, ASP.NET 2.0 sets the Cache-control to private because max-age = 0 means no cache is needed. So, there's no way you can make ASP.NET 2.0 return proper headers which caches the response.
Time for a hack. After decompiling the code of the HttpCachePolicy class (Context.Response.Cache object's class), I found the following code:
Somehow, this._maxAge is getting set to zero and the check: "if (!this._isMaxAgeSet || (delta < this._maxAge))" is preventing it from getting set to a bigger value. Due to this problem, we need to bypass the SetMaxAge function and set the value of the _maxAge field directly, using Reflection.
[WebMethod][ScriptMethod(UseHttpGet=true)]
public string CachedGet2()
{
TimeSpan cacheDuration = TimeSpan.FromMinutes(1);
FieldInfo maxAge = Context.Response.Cache.GetType().GetField("_maxAge",
BindingFlags.Instance|BindingFlags.NonPublic);
maxAge.SetValue(Context.Response.Cache, cacheDuration);
Context.Response.Cache.SetCacheability(HttpCacheability.Public);
Context.Response.Cache.SetExpires(DateTime.Now.Add(cacheDuration));
Context.Response.Cache.AppendCacheExtension(
"must-revalidate, proxy-revalidate");
return DateTime.Now.ToString();
}
This will return the following headers:
Now max-age is set to 60 and thus the browser will cache the response for 60 seconds. If you make the same call again within 60 seconds, it will return the same response. Here's a test output which shows the date time returned from the server:
After 1 minute, the cache expires and the browser makes a call to the server again. The client side code is like this:
function testCache()
{
TestService.CachedGet(function(result)
{
debug.trace(result);
});
}
When 'this' is not really 'this'
Atlas callbacks are not executed on the same context where they are called. For example, if you are making a Web method call from a JavaScript class like this:
function SampleClass()
{
this.id = 1;
this.call = function()
{
TestService.DoSomething( "Hi", function(result)
{
debug.dump( this.id );
} );
}
}
What happens when you call the "call" method? Do you get "1" on the debug console? No, you get "null" on the debug console because "this" is no longer the instance of the class. This is a common mistake everyone makes. As this is not yet documented in Atlas documentations, I have seen many developers spend time finding out what's wrong.
Here's the reason. We know whenever JavaScript events are raised, "this" refers to the HTML element which produced the event. So, if you do this:
function SampleClass()
{
this.id = 1;
this.call = function()
{
TestService.DoSomething( "Hi", function(result)
{
debug.dump( this.id );
} );
}
}
<input type="button" id="ButtonID" οnclick="o.onclick" />
If you click the button, you see "ButtonID" instead of "1". The reason is that, the button is making the call. So, the call is made within the button object's context and thus "this" maps to the button object.
Similarly, when XML HTTP raises the event onreadystatechanged which Atlas traps and fires the callback, the code execution is still on the XML HTTP's context. It's the XML HTTP object which raises the event. As a result, "this" refers to the XML HTTP object, not your own class where the callback is declared.
In order to make the callback fire on the context of the instance of the class so that "this" refers to the instance of the class, you need to make the following change:
function SampleClass()
{
this.id = 1;
this.call = function()
{
TestService.DoSomething( "Hi",
Function.createDelegate( this, function(result)
{
debug.dump( this.id );
} ) );
}
}
Here, the Function.createDelegate is used to create a delegate which calls the given function under the "this" context. Function.createDelegate is defined in AtlasRuntime:
Function.createDelegate = function(instance, method) {
return function() {
return method.apply(instance, arguments);
}
}
HTTP POST is slower than HTTP GET but it is default in ASP.NET AJAX
ASP.NET AJAX, by default, makes HTTP POST for all web service calls. HTTP POST is more expensive than HTTP GET. It transmits more bytes over the wire, thus taking precious network time, and it also makes ASP.NET do extra processing on the server end. So, you should use HTTP GET as much as possible. However, HTTP GET does not allow you to pass objects as parameters. You can pass numerics, string, and date only. When you make an HTTP GET call, Atlas builds an encoded URL and makes a hit to that URL. So, you must not pass too much content which makes the URL become larger than 2048 chars. As far as I know, that's what is the max length of any URL.
In order to enable HTTP GET on a web service method, you need to decorate the method with the [ScriptMethod(UseHttpGet=true)] attribute:
[WebMethod] [ScriptMethod(UseHttpGet=true)]
public string HelloWorld()
{
}
Another problem of POST vs. GET is, POST makes two network roundtrips. When you first make a POST, the web server sends an "HTTP 100 Continue" which means that the web server is ready to accept the content. After that, the browser sends the actual data. So, initiation of a POST request takes more time than GET. Network latency (roundtrip time between your computer and the server) is the biggest concern in AJAX applications because AJAX makes many small calls which needs to be done within milliseconds. Otherwise the application does not feel smooth and creates user annoyance.
Ethereal is a nice tool to see what happens under the hood on POST and GET:
From the above picture, you see that POST requires a confirmation from the web server: "HTTP/1.1 100 Continue" before sending the actual data. After that, it transmits the data. On the other hand, GET transmits the data without waiting for any confirmation.
So, you should use HTTP GET while downloading data from a server like parts of pages, contents in a grid, or a block of text etc. But you should not use HTTP GET to send data to a server like submission of web forms.
Conclusion
The above extreme hacks are implemented in Pageflakes, not in the exact way as mentioned here, but the principles are the same. So, you can happily rely on these techniques. These techniques will save you from many problems that you will probably never realize in your development environment, but people from all over the world will face those problems when you will go on large scale production. Having these tricks implemented right from the beginning will save you a lot of development and customer support effort. Keep an eye on my blog for more tricks to come.
