We recently tested an umbraco instalation that called thrid part webservices and reformatted responces USING browsermob. 50 concurent users hitting 9000 times in 30 minuites (non cashed pages). Whilst responce time dropped it only went doen to 20 seconds wich for the machine was quite nice me thinks.
Thanks for sharing. Why do you feel that a response time of 20 seconds (for a single request!?) is nice? Anybody visiting your website is gone after about 5 seconds of non-responsiveness (if not sooner).
What is acceptable performance under high load is always subjective, though 20 seconds is not fast to be sure.
But do notice a few things about this test...
9000 hits in 30 minutes = 300 hits/min = 425 thousand hits/day or 12 million hits/month - if each of the 50 users had this activity level in the test that would be more than half a billion hits/month - but I've been conservative because the Alec didn't specify if 9000 were total or per concurrent user
this was (apparantly) a single web server and umbraco can be load balanced easily for scalability
umbraco did not die or give errors
the site remained responsive (though sluggish)
macro caching and other performance aids would help further
it may be that the webservice being called added to the overall slowness as load increased on it as well
I'd say this was a wonderful success. With caching, looking for performance bottlenecks in macros (and possibly the webservice), and the ability to easily add multiple servers behind a load balancer the sky is the limit for performance with umbraco.
Hi. I should have given more information on the test and may have made a mistake with the numbers, i will check tomorrow.
Basicly, the umb macro called another (in company) .Net based webservice that we maintain. The responce of which on the staging server can be 1-2 seconds alone (and this is probably a big part of the reason for the 20 seconds). We wanted to do a single test that would test both systems without caching just to try and kill the server using a trial account for the online service. We used the free browsermob account that woerks with a limited number of credits and only allows 50 concurent virtual users (it can go to 1000's if you pay), The system uses a system of Java test test scripts that can do multiple things like checking responces, times and log to a db. This is the reason the test was not so sientific. We have tried client based load testing services but you alway end up testing the client/network not the server firewall gateway.
I am very happy with umb and the service mentioned and after we move our server later this year I will get my wallet out and see how far we can push the system again.
Umbraco prformance test results
Just FYI
We recently tested an umbraco instalation that called thrid part webservices and reformatted responces USING browsermob. 50 concurent users hitting 9000 times in 30 minuites (non cashed pages). Whilst responce time dropped it only went doen to 20 seconds wich for the machine was quite nice me thinks.
8gb ram, umb 4.0x, seperate db server
Thanks for sharing. Why do you feel that a response time of 20 seconds (for a single request!?) is nice? Anybody visiting your website is gone after about 5 seconds of non-responsiveness (if not sooner).
What is acceptable performance under high load is always subjective, though 20 seconds is not fast to be sure.
But do notice a few things about this test...
- if each of the 50 users had this activity level in the test that would be more than half a billion hits/month
- but I've been conservative because the Alec didn't specify if 9000 were total or per concurrent user
I'd say this was a wonderful success. With caching, looking for performance bottlenecks in macros (and possibly the webservice), and the ability to easily add multiple servers behind a load balancer the sky is the limit for performance with umbraco.
cheers, and thanks for sharing, Alec!
doug.
Hi. I should have given more information on the test and may have made a mistake with the numbers, i will check tomorrow.
Basicly, the umb macro called another (in company) .Net based webservice that we maintain. The responce of which on the staging server can be 1-2 seconds alone (and this is probably a big part of the reason for the 20 seconds). We wanted to do a single test that would test both systems without caching just to try and kill the server using a trial account for the online service. We used the free browsermob account that woerks with a limited number of credits and only allows 50 concurent virtual users (it can go to 1000's if you pay), The system uses a system of Java test test scripts that can do multiple things like checking responces, times and log to a db. This is the reason the test was not so sientific. We have tried client based load testing services but you alway end up testing the client/network not the server firewall gateway.
I am very happy with umb and the service mentioned and after we move our server later this year I will get my wallet out and see how far we can push the system again.
is working on a reply...