Sater Posted February 1, 2015 Posted February 1, 2015 Hi, I want to pay the community and, in the first order, the development team attention that the 4.0 looks has issues with the performance, comparing to previous versions. I did these tests to make sure if I will upgrade to 4.0 my production, this will not require any efforts and investments for my current hosting. BTW, currently I'm using 2.1.7 and it's time to move forward. What I did, I run 10 instances of the Chrome, as used an auto refresh plugin, with schedule every 10 seconds. Pages reloaded with random start. So, in general, it simulated 10 active reading users on my board, where each of them click a link every 10 seconds randomly - in average 1 click/sec. First I run this test for 4.0 Beta 5. And see the graph below - it almost reached a situation, when a hosting would block my account. My limit is 50% of CPU avarage per month. When a double limit crossed - my account would be blocked. Not really an issue - but would be very surprising to shut down my production for few hours. Hopefully, I was not able to run the tests for whole day. Only about 10 hours, plus - minus. Then I decided to roll back to 3.4.7 to see what is going there. See the results also below. At the first glance, it could look, like 4.0 B5 doesn't have issues, if to analyze number of loaded pages only. BUT, if you will calculate the load I created (~10 hours * 3600 clicks) and also look on the traffic it generated, it becames obvious - that to load a single page for a user, the IPB 4.0 requesting few pages from the hosting. And this creates enormous load on the hosting with the same number of visitors. From another side, I see 3.4.7 behaves very good. It doesn't require much more resources, comparing to 8 years old 2.1.7. What is just great! P.S. The pages I used for analysis to refresh every 10 sec. are most used - main, list of topics in a forum, different topics, e.t.c. Mean, most common user behavior, who read a forum. P.S. Apologize for non English statistic - this is only what offers my hosting provider.
AndyF Posted February 1, 2015 Posted February 1, 2015 Unsure if any comparision with the 2x series is fair though as the 2x and 1x series were lightweight without many features so they are going to be faster loading than 3 or 4... Interesting read though.
Management Charles Posted February 1, 2015 Management Posted February 1, 2015 Performance in 4.0, as in 3.4, is something we continually improve with each release. The 4.0 you benchmarked is a lot better than the 4.0 of a few weeks or a few months ago. You may wish to enable MySQL datastore method rather than the default file system datastore if your main concern is memory/CPU usage. Please try that first and then wait until RC 1 and/or 4.0.0 final and see if there are improvements. Look forward to you reporting back
Sater Posted February 1, 2015 Author Posted February 1, 2015 Performance in 4.0, as in 3.4, is something we continually improve with each release. The 4.0 you benchmarked is a lot better than the 4.0 of a few weeks or a few months ago. You may wish to enable MySQL datastore method rather than the default file system datastore if your main concern is memory/CPU usage. Please try that first and then wait until RC 1 and/or 4.0.0 final and see if there are improvements. Look forward to you reporting back Yes. I have a plan to test the performance with caching enabled and, of-course, with other further versions of 4.0 before to make the clue and decision. Will keep you updated! Just want to make sure the IPS team is aware about issues before go to the release.
Management Charles Posted February 1, 2015 Management Posted February 1, 2015 Before you enable caching just try switching from file system to MySQL datastore. It's just a toggle under Advanced Configuration in AdminCP.Caching is great but it can often just mask things.
bfarber Posted February 2, 2015 Posted February 2, 2015 Is there any way you can translate that last image for me? I get the general gist of it (load testing against 4.0 showed a spike compared to 2.x and 3.x) but I'm not clear on what each line actually says. For instance, if 4.0 served 1.86GB over 220,000 requests and 3.4 served 2.7GB over 72,000 requests, then (1) 4.0 would save a lot of bandwidth (which would help with performance undoubtedly) and (2) while 4.0 spiked higher than 3.4, it also served about 3x more requests. That said, I'm not actually sure that's what the graph says, and I don't want to make any assumptions.
Alexander V Posted February 2, 2015 Posted February 2, 2015 Topleft: "Used resources http servers", "Schedule Processor Load", and again Schedule Processor Load on the right. Used Google's new camera translation tool
Management Charles Posted February 2, 2015 Management Posted February 2, 2015 @Sater Please do follow up once you change the datastore method from file system to MySQL. I am curious what you see. Thanks!
Sater Posted February 3, 2015 Author Posted February 3, 2015 Is there any way you can translate that last image for me? I get the general gist of it (load testing against 4.0 showed a spike compared to 2.x and 3.x) but I'm not clear on what each line actually says. For instance, if 4.0 served 1.86GB over 220,000 requests and 3.4 served 2.7GB over 72,000 requests, then (1) 4.0 would save a lot of bandwidth (which would help with performance undoubtedly) and (2) while 4.0 spiked higher than 3.4, it also served about 3x more requests. That said, I'm not actually sure that's what the graph says, and I don't want to make any assumptions. That the issue. Because the statistic shows over 200 000 pages loaded from the server, while "user", actually, requested only about 35 000 pages. P.S. And guys, it's so lacking feature - quote selected text in IPS. Don't understand why it's still not in the product out of the box.
Recommended Posts
Archived
This topic is now archived and is closed to further replies.