Posted by ClickInfluence on September 7, 2010 at 1:13pm
Hi all, I'm sure there must have been some tests run on Pantheon, but I can't find any documentation of such anywhere.
Specifically I'm interested to learn how many logged in users (doing regular stuff like browse the site, post a question, look at a profile etc) could pantheon running on an EC2 server created from the AMI handle?
At what level of logged in users might we expect Pantheon to start having trouble?
Comments
That is going to be
That is going to be impossible to say. Even with the exact same Pantheon base and hardware platform, there will be huge differences with different modules, number of dynamic blocks on a page, query optimization, number of nodes, number of users, etc. One slow module or complex view can make a huge difference.
There probably needs to be
There probably needs to be some kind of baseline to work from then. Say a default install with x,y and z core modules enabled and a few core blocks. Anything would be good. All I need (and I'm guessing this would be useful for a whole stack of folks) is something to base my guess on -- a point at which to begin...
re: There probably needs to be
The other issue is that VPS servers can vary from server-to-server and day-to-day - maybe EC2 more than most.
Greg
--
Greg Coit
Systems Administrator
http://www.chapterthree.com
Yes
We have been getting a review from webperformanceinc.com which will give some third-party (e.g. neutral) stats.
However, what this means definatively for your site and your hosting is impossible to say given all the variations that can happen within those. Benchmarks are just benchmarks. ;)
https://pantheon.io | http://www.chapterthree.com | https://www.outlandishjosh.com
But still..
I can see your point and you are more that right, Josh and Greg.
But I also see the point in ClickInfluence question so I'll try to rephrase the question:
Let me say I am on some hosting plan X where I bulid my site with these modules, themes, hooks, views, panels and all other things (follows insane list of everything Drupal). Now I feel that performance is maybe sluggish and I am considering moving to EC with Pantheon, questions are:
Now I will try to sound more knowledgeable.
If site is not even built but person is just asking, then it is easy: go for as much as you can afford. Silly as it does sound, actually it is true: your own server (provided you have time and knowledge to not only maintain but also to optimize every aspect) will give you fast and responsive site, coupled with all Pantheon bells and whistles, you'll have something you could easily be proud of. Of course, that power does not justify lousy site architecture or crappy modules but it will take time (and load) for that to come out.
Now, if one does have a site, this is possible scenario (or question). You can buy EC for a day just to test drive how good it will be. So, you have your EC AMI running, setup your site, put all modules, themes, hooks, views, panels, etc. Now you have ye old hosting working as it is and new beast ready with replicated site.
The question is: how can I stress load them in order to see how each will behave under load? How many users can be logger at same time or how many can actually request some content and get served before some dreaded errors start to pop up or server falls on its knees and beg for mercy?
We do have small little tool called
abthat could benchmark our site (both of them) but what would be the best scenario (parameters to use withab) and could this tool give us good information?This is kind of information I keep asking myself and also I believe many other beginners and that is why I think this kind of information should be put in some kind of documentation and FAQ. Many will be grateful and this question will not be repeated over again.
And the best answer can come only from source ;)
_______________
Use Mac.
Mladen Đurić
Web Development.
But still..
Ab can be used to do basic drupal performance testing, but for more robust and informative testing, you'll want something else. We've done some work with jmeter but there's more that can/will be done.
BTW, the test that joshk mentioned appears to have been posted online:
Part 1: http://www.webperformanceinc.com/load_testing/blog/2010/09/optimizing-drupal-introduction/
Part 2: http://www.webperformanceinc.com/load_testing/blog/2010/09/optimizing-drupal-from-baseline-drupal-to-the-pantheon-drupal-platform/
Greg
--
Greg Coit
Systems Administrator
http://www.chapterthree.com