37 Million, 9 hours.

I had to share this with anyone who would listen because this is so friggin cool. I’ve been working on this project at work and we’re using Flash Remoting for services, and we are expecting millions of hits, the most being 7 million a day, but averaging in the hundreds of thousands of users per day.

So we had to devise a way to stress test the server software. That’s the screen capture I posted a couple days ago. But anyway, I setup 7 computers that all had JMeter running – hitting our internal dev server. Total there were 2000 threads, sending 28 requests per thread. In 9 hours, the server handled 37 million requests without anything being a problem. That’s crazy!

The other thing that was so crazy is that each request or “transaction” hits the MS SQL server 5 times per transaction. The server side software is awesome. Chris Crissmond and Darrell Thurmond had written it, its all free threaded – basically one transaction takes 0 seconds. It’s written in C# and SQL Server. These guys did an amazing job.

Another thing that happens per transaction is hitting the M$ Live Search API. M$ themselves told us we wouldn’t be able to get more than about 40 transactions per second on hardware we had (32 bit, 3 gigs ram, dual core something). But we peaked at processing about 300 requests per second. The server we had didn’t even care about what was happening, memory was at about 540MB, and the cpu was staying between 25-40% usage.

The next morning, we had the Live Search team from M$ asking us how the hell we did that. Pretty friggin cool. Also note that JMeter bursts requests, it’s not constantly sending requests. So we probably could have done 90 million in 9 hours no problem. Crazy.

Another successful AMF project. Go AMF!

Required said,

January 29, 2008 @ 00:28

What’s this got to do with AMF? Sounds like you’re trying to sell m$ software.

aaron said,

January 29, 2008 @ 00:34

haha. this has nothing to do with M$. They were just project requirements. I’m just trying to share something that I though was extremely cool. I could care less about what this was done with. It’s just cool man.

aaron said,

January 29, 2008 @ 00:36

As far as being related to AMF. I used all the AMF stress testing tools I had documented. And AMF is fast!

AG said,

January 29, 2008 @ 21:24

Is http://wiki.rubyamf.org/ broken?

aaron said,

January 29, 2008 @ 22:10

yeah it isn’t the source for docs anymore. I’ve been meaning to take it down actually. Sorry :( .

RSS feed for comments on this post · TrackBack URI

Leave a Comment