How scalable is askbot?
Could it run on really large websites, like Stackoverflow..
What are the biggest sites?
1 Answer
- Sort by
oldest newest most voted
5
To date (Dec 2011) the biggest public site is http://answers.ros.org and probably the second biggest http://ask.fedoraproject.org.
It seems that on stackoverflow there are several content writes per minute - in single digits - a single database and an application server will handle that, askbot can do that too - easily.
According to this post (published in March 2011) - http://highscalability.com/blog/2011/3/3/stack-overflow-architecture-update-now-at-95-million-page-vi.html - stackoverflow takes about 100 million pageviews per month - about 40 pageviews per second on average.
Askbot takes 200 ms to generate some pages without caching - that gives only 4 page views per second (on heavy pages, probably 6 if you count in statistics of various page loads). So you'll need 10 read servers working in parallel and one master database server to match performance of stackoverflow with askbot. Possible, but you will need resources and a sys admin.
We are working on serious caching support and the numbers will improve significantly.
Right now I think the pressure from the users is to better support concurrent reads - so we will solve this problem first. For concurrent writes - we'll think of that when some client proves that they actually have that problem.
Comments
Thank you, Evgeny! good answer!
Aren't these numbers too optimistic? I tried pingdom and other tools (webpagetest): common pages (such as this page, or the main /questions/) always start rendering >1s, and a complete load time is usually above 3s.
Actually the numbers I listed are measured on the server side - the time to render page by the server. The webpagetest will measure entire round trip between the web browser and the server. Try answers.ros.org or ask.sagemath.org those are faster. Also, you should only hit the complete load time once - to load all the js, which loads at the bottom of the page.
here comparison between answers.ros.org : http://www.webpagetest.org/result/111216XC2HXWW/1/screenshot/ and stackoverflow: http://www.webpagetest.org/result/111216852HY6Y/1/screenshot/ stack seems to much faster
Interesting discussion! :) I don't think speed is very important at the moment, but for the record, webpagetest measures both "first view" and "repeat view", and breaks down the times for different activities/resources -- so it's not just "full load time for uncached view".
Comments
To answer this we'll need to look at Stackoverflow numbers, we'll try to look up tomorrow.