According to this (http://news.yahoo.com/s/nm/20070820/media_nm/myspace_dc_2) MySpace is at about 61 million active, unique users while Facebook is at about 19 million. Also MySpace users spend more time on the site than Facebook uses, which I'm sure translates to more traffic per user.
As a very unscientific test, I just logged on to my Facebook account and my Myspace account and noted the following about my home pages:
Myspace
html size: 16.38kb
count of external .js files: 10
count of external .css files: 4
render mode: quirks
load time (time to reload, thus assuming most requests cached, average of 3): 3191ms
Facebook
html size: 5.35kb
count of external .js files: 27
count of external .css files: 25
render mode: standards
load time (time to reload, thus assuming most requests cached, average of 3): 3207ms
I was surprised the load times were so close. MySpace feels slower but that may be because pages frequently include external images and sound files.
I don't mean these comparisions to either trash or promote MySpace or Facebook; instead I think it's instructive to consider which benchmarks translate to the best user experience and what techniques contribute to the better set of numbers.
You're about two years too late to do benchmarks on myspace as a startup. They've re-written the thing at least twice over and put a whole lot more money into fixing it up by now.
Myspace's server software and html was bad. So so bad.
I'm guessing your estimates are correct enough. I doubt the user count for all YC projects exceeds 500,000, and according to http://forevergeek.com/articles/debunking_the_myspace_myth_o..., MySpace has at least 43 million, which by your math would mean that Facebook has around 14 million users. Having been a frequent MySpace user myself in the not-too-distant past (I'm 22, and I used it from age 18-20), I concur that there have been times in which MySpace "limped along". It was not uncommon to get multiple "application error" (for lack of a better term -- I don't know/remember what it's called under .NET) during a surfing session, or to encounter a page or profile that just would not load. I've since switched to Facebook, and in comparison, I've never had Facebook barf on me. This, to me, evidences either the dim-wittedness of MySpace's developers, or superiority of Facebook's web solution (http://article.gmane.org/gmane.comp.web.cache.memcached/3224) over what MySpace has chosen.
Note that I am not saying that PHP > .NET -- obviously on sites of this large scale, as has already been noted in the post's comments, using any technology "out of the box" is looking for trouble. What I mean to imply, however, is that perhaps for sites this big, .NET is the wrong thing. And, as the parent comment points, out, IIS certainly isn't helping things, though I don't think Apache would necessarily be a magic bullet. Craigslist, which does many megabytes of data transfer per minute, is considering switching to Lighttpd because Apache too heavyweight.
P.S. Sorry about the links -- I couldn't figure out how to link a text string to a URL. Can someone enlighten me?
Note that .net is essentially just a VM and set of APIs. There is nothing stopping someone from building site to service massive loads. The bottlenecks occur when .net is paired up with IIS, and various asp.net controls and ViewState.
For example, we use apache with mod_mono and write our own custom controls.
Couldn't the same argument ("just a VM and a set of APIs") be made for many programming languages, Web or otherwise? I'm not criticizing what you wrote, just trying to make sure I understand what you're saying.
Gotcha. I'm just arguing (mildly, 'cause the most .NET experience I have is a two-week project a couple years ago) that possibly .NET and friends is not the best choice for a massive site such as MySpace, at least if you are not prepared/qualified to completely reinvent the wheel (as you seem to have done). Then again, Facebook's solution seems pretty ad-hoc as well. Perhaps such wheel invention is inevitable.
Yeah I see your points. I just wanted to point out that ".net" is different than asp.net and IIS. Those are part of the complete default stack and are typical bottlenecks. In a way they're designed to achive easy development/deployment over performance.
asp.net out-of-the-box will not work for a massively concurrent public facing app. However, neither will php, rails, django, any of the java frameworks. "re-invent the wheel" might be an overstatement for what you need to do, but no matter what...you're going to do some serious tweaking since every app has different needs
