Comments on: Load/performance testing a Rails application with ApacheBench http://www.yes-no-cancel.co.uk/2008/10/27/load-performance-testing-a-rails-application-with-apachebench/ Entrepreneurship, web technology and the user experience Tue, 23 Feb 2010 19:18:22 +0000 http://wordpress.org/?v=2.9.2 hourly 1 By: Martin Kleppmann http://www.yes-no-cancel.co.uk/2008/10/27/load-performance-testing-a-rails-application-with-apachebench/comment-page-1/#comment-1012 Martin Kleppmann Mon, 03 Nov 2008 17:36:26 +0000 http://www.yes-no-cancel.co.uk/?p=166#comment-1012 Tom: A very good point. This is something we thought about beforehand, and we went for a compromise. The thing with a site which relies on providing a community or a marketplace, like this, is that you need to get a certain base level of usage, otherwise the site is useless to even the most good-willing early adopters. If you visit an auction site and don't see anything for sale, you're certainly not going to be the first person to list something, because most likely it won't be worth your time. The 'network externality' problem. So we had to draw enough attention to the site prior to launch to allow us to pre-fill it at least with two dozen lots, enough to make the site look alive; on the other hand, that's still a small enough number that it was very unlikely to take our servers down on launch. Regarding the magazine articles, we timed the launch so that it happened before the publication of the article (the launch spike would already have passed), but before the major auctions close (we hopefully receive lots of bids from interested buyers). We're still watching the stats, but so far it seems like this strategy is working out nicely. Tom: A very good point. This is something we thought about beforehand, and we went for a compromise. The thing with a site which relies on providing a community or a marketplace, like this, is that you need to get a certain base level of usage, otherwise the site is useless to even the most good-willing early adopters. If you visit an auction site and don’t see anything for sale, you’re certainly not going to be the first person to list something, because most likely it won’t be worth your time. The ‘network externality’ problem.

So we had to draw enough attention to the site prior to launch to allow us to pre-fill it at least with two dozen lots, enough to make the site look alive; on the other hand, that’s still a small enough number that it was very unlikely to take our servers down on launch.

Regarding the magazine articles, we timed the launch so that it happened before the publication of the article (the launch spike would already have passed), but before the major auctions close (we hopefully receive lots of bids from interested buyers).

We’re still watching the stats, but so far it seems like this strategy is working out nicely.

]]> By: Tom Cashman http://www.yes-no-cancel.co.uk/2008/10/27/load-performance-testing-a-rails-application-with-apachebench/comment-page-1/#comment-963 Tom Cashman Tue, 28 Oct 2008 09:58:03 +0000 http://www.yes-no-cancel.co.uk/?p=166#comment-963 I'm guilty here of responding to the first paragraph, rather than the comprehensive load testing that this post's actually about. But I'm interested... did you consider other launch models? The "big bang launch" strategy received a bit of criticism recently when new search engine "Cuil" went under on day one. Some bloggers thought they should have used a gmail-like "slow burn" start instead (e.g. http://www.aaronsw.com/weblog/howtolaunch ). Any thoughts? I’m guilty here of responding to the first paragraph, rather than the comprehensive load testing that this post’s actually about. But I’m interested… did you consider other launch models? The “big bang launch” strategy received a bit of criticism recently when new search engine “Cuil” went under on day one. Some bloggers thought they should have used a gmail-like “slow burn” start instead (e.g. http://www.aaronsw.com/weblog/howtolaunch ). Any thoughts?

]]>