Hi all, I'm hosting a drupal site on a dedicated server amd64 and 8Gb RAM for 5000 users/day, most of them anonymous (it is a news site so only journalist has to log in). I've cached it with boost module.
I also use apache2 and php5-cgi with apc, mysql query caching...
So as the end, the site should fly, but it is not so performing as you can see http://tinyurl.com/64fyh2d
The speed may change randomly.. sometimes it's very slow other times it's quite good, and it seems not depending on the number of visitors.
Ah, i don't use cache on views because they take a long time to update the content, and the site is updated very very often (every 5/10 minutes a new entry or a change). With caching on view actived the content are not displayed in real time.

Can you help me optimizing it?

Comments

GDrupal’s picture

The real professional approach to boost you site along with a good server configuration:

http://www.dynamiteheads.com/blog/jakub-suchy/migrating-your-site-pressf...

----------------------------------------------------------
G.A. Martin, Drupal Developer.

Cosmy’s picture

Thank you for your reply.
No, I'm not using it, but I think that at the moment the site should be faster, because 5000 user a day are not so much to slow up a site with cache and up to date dedicated hardware.
I'll consider an architectural migration only when the traffic will justify it, otherwise the risk to export some misworking element.

I've noticed in the system monitor (top) a lot oh php5 proesses. I think there is something wrong in the code, but i don't know what, and i don't know how to find where the problem is. I've also to say that update and insert operations are very very slow. Sometimes, in the worst cases, they take about a minute. I think it's cache generation matter, but i can't disable cache to test it..

GDrupal’s picture

You can run a local test but in most case a pressflow core migration is transparent... all will works just the same... but faster... it has disabled some php4 compatibilities and other things that if you are running an apache with php5 will not cause you any problems...

----------------------------------------------------------
G.A. Martin, Drupal Developer.

Cosmy’s picture

I've installed pressflow and migrated to innodb, and of course, for a bit my statistics raised up as the article explains.
Now, after 10 days, I'm registering worst statistics (of course, not so bad as before installing pressflow), but there are slow queries again (the first days there weren't slow queries at all!). They are all operations (INSERT, SELECT and DELETE) in drupal_boost_cache tables..
so i think there is something wrong between pressflow and boost module.

Do you know something about this issue?

GDrupal’s picture

Check this link, and tell me if you find something but I think that might be something else. It will require a deep analysis, can you post apache logs, reports, snapshots, statistics?.

http://drupal.org/project/issues/boost?text=pressflow&status=All

I just check your site It seems to me that it is flying like the wind :-P, perhaps there are times with large traffic than anothers.

Finally check this too.... will definitely help you a lot:

http://groups.drupal.org/node/15663

Let me know how it goes...

----------------------------------------------------------
G.A. Martin, Drupal Developer.

Cosmy’s picture

thank you for your reply.
After that posts I've made my self hard working and doing several tests with different configurations. I've also changed the graphic and done a deploy system with minified css and gzip policies. So what you have seen is a new version.

So, back to the topic, I've found few critical things (pre 16th may):
1. Anonymous users wasn't using the cache. In fact the node access counter was working also for anonymous and it shouldn't when seeing cached pages. I don't know the real cause of this, maybe a problem with page cache expiration (I suspect that after expiration they weren't regenerated).
2. Despite this Boost module was operating and adding rows and rows in DB, so teh insert and update operations were very very slow, it generated the cache but the users weren't using it really.
3. Cache DB was huge
4. It wasn't serving contents through gzip.

So i've tried to disable boost and noticed that the site was very performing.
Then after truncating all boost_cache tables I've done this things:
1. disabled core cache and gzip compression (I have mod_deflate enabled)
2. checked some parameter to Boost enabling this functions:
Check database timestamps for any site changes. Only if there has been a change will boost flush the expired content on cron
3. installed rules module and set it up for running cron everytime a change or an insert is done. It makes the insert process a little bit long (from 10 to 30 seconds not more), but not never ending like before.

As you noticed in your message, now the site is faster. But Yslow grade is C, so it's not so fast as i'm trying to do. When i disable Boost i reach grade B (very very strange), but i need boost to serve peaks and high traffic moments.
I hope now the cache DB won't get huge another time

GDrupal’s picture

Good to know that you are more close to the solution and thanks for the details... It will surely help to someone in the future! Personally I will soon begin with a huge migration for a newspaper site that has 40.000 request per hour, :-x so, I will go deep in the performance issues, wish me luck!

----------------------------------------------------------
G.A. Martin, Drupal Developer.

ayesh’s picture

Don't blame on Boost, please!
This mostly happen when your boost configuration. Tune it up to flush caches in an intelligent manner. For an example, if your site has number of nodes, flushing cache when node insert is the worst! There are number of ways to flush caches especiually on high-traffic(if you say 5000 visitors per day is higher) using Rules and Action/triggers.
I have read many posts in this forum(forum/22) saying boost made site slow. Most of them got it solved by setting up correct Boost caches.

Also, do not try to boost too much. I have bad experience when caching XML files when I don't actually provide feeds. Also, XML Sitemap module keeps static files and in most cases no need to cache it.

Use Pressflow's path alias performance extension. It works really well!
The only drawback in Pressflow is, it "advertises" that it's pressflow!