ASP.NET: Legitimate architecture/HttpModule concern?

Question

An architect at my work recently read Yahoo!'s Exceptional Performance Best Practices guide where it says to use a far-future Expires header for resources used by a page such as JavaScript, CSS, and images. The idea is you set a Expires header for these resources years into the future so they're always cached by the browser, and whenever we change the file and therefore need the browser to request the resource again instead of using its cache, change the filename by adding a version number.

Instead of incorporating this into our build process though, he has another idea. Instead of changing file names in source and on the server disk for each build (granted, that would be tedious), we're going to fake it. His plan is to set far-future expires on said resources, then implement two HttpModules.

One module will intercept all the Response streams of our ASPX and HTML pages before they go out, look for resource links and tack on a version parameter that is the file's last modified date. The other HttpModule will handle all requests for resources and simply ignore the version portion of the address. That way, the browser always requests a new resource file each time it has changed on disk, without ever actually having to change the name of the file on disk.

Make sense?

My concern relates to the module that rewrites the ASPX/HTML page Response stream. He's simply going to apply a bunch of Regex.Replace() on "src" attributes of <script> and <img> tags, and "href" attribute of <link> tags. This is going to happen for every single request on the server whose content type is "text/html." Potentially hundreds or thousands a minute.

I understand that HttpModules are hooked into the IIS pipeline, but this has got to add a prohibitive delay in the time it takes IIS to send out HTTP responses. No? What do you think?

Answer 1

He's worried about stuff not being cached on the client - obviously this depends somewhat on how the user population has their browsers configured; if it's the default config then I doubt you'd need to worry about trying to second guess the client caching, it's too hard and the results aren't guaranteed, also it's not going to help new users.

As far as the HTTP Modules go - in principle I would say they are fine, but you'll want them to be blindingly fast and efficient if you take that track; it's probably worth trying out. I can't speak on the appropriateness of use RegEx to do what you want done inside, though.

If you're looking for high performance, I suggest you (or your architect) do some reading (and I don't mean that in a nasty way). I learnt something recently which I think will help -let me explain (and maybe you guys know this already).

Browsers only hold a limited number of simultaneous connections open to a specific hostname at any one time. e.g, IE6 will only do 6 connections to say www.foo.net.

If you call your images from say images.foo.net you get 6 new connections straight away. The idea is to seperate out different content types into different hostnames (css.foo.net, scripts.foo.net, ajaxcalls.foo.net) that way you'll be making sure the browser is really working on your behalf.

Answer 2

A few things to be aware of:

1. If the idea is to add a query string to the static file names to indicate their version, unfortunately that will also prevent caching by the kernel-mode HTTP driver (http.sys)
2. Scanning each entire response based on a bunch of regular expressions will be slow, slow, slow. It's also likely to be unreliable, with hard-to-predict corner cases.

A few alternatives:

1. Use control adapters to explicitly replace certain URLs or paths with the current version. That allows you to focus specifically on images, CSS, etc.
2. Change folder names instead of file names when you version static files
3. Consider using ASP.NET skins to help centralize file names. That will help simplify maintenance.

In case it's helpful, I cover this subject in my book (Ultra-Fast ASP.NET), including code examples.

Answer 3

http://code.google.com/p/talifun-web

StaticFileHandler - Serve Static Files in a cachable, resumable way. CrusherModule - Serve compressed versioned JS and CSS in a cachable way.

You don't quite get kernel mode caching speed but serving from HttpRuntime.Cache has its advantages. Kernel Mode cache can't cache partial responses and you don't have fine grained control of the cache. The most important thing to implement is a consistent etag header and expires header. This will improve your site performance more than anything else.

Reducing the number of files served is probably one of the best ways to improve the speed of your website. The CrusherModule combines all the css on your site into one file and all the js into another file.

Memory is cheap, hard drives are slow, so use it!