Moving from lighttpd to Apache 2.2
A while ago, I blogged about our shift from apache to lighttpd as our production webserver.
Lighttpd has been good for us, but we’ve been experience a number of the proxying problems with lighttpd. To make matters worse, I’ve been unhappy with the lack of production quality dynamic httpd compression. With the new zoomin its provided an opportunity to test apache 2.2 and mongrel in real world.
I have been testing mongrel and apache 2.2 for a week on zoomin production and I’ve observed few of things.
- Mongrel seems to be fractionally slower than fastcgi / lighttpd. For one of our pages its 25 pages per second to 27 pages per second.
- Apache Worker MPM module is a lot more effective than apache prefork MPM (default MPM module) in unix process management and performance. My performance tests show the worker module almost as fast as lighttpd in delivering static content. We have been able to deliver static content at 2700 req/sec. I am watching with interesting the apache’s new event MPM.
- Using apache / mongrel and compression, the total bytes transferred has decreased by 60% and search indexing by google has increased by 20%.
- Pages are snappier downloading. Our biggest page – New Zealand Place page(which is 98k compresses to 16k ) and pops back very quickly.
We’ve been very happy we the transition back to apache and I’ve been recommending it to other rails shops around town. I’ll be sticking to lighttpd to deliver our static content and I’m looking with interest at lighttpd 1.5 and will be put it up for a pepsi challenge versus apache 2.2 as soon as it hits production.