Are we reaching traffic limit of IPB ?

[Lauren3]

We actually run our own dedicated server for IPB. Our traffic is most of time over 700 connections for last 60 mn, and during frequent peak periods, we have 1000+ connections.
Our server is quite robust, 2.8GHz Quad-Core Intel Xeon loaded with 14Gb RAM on Mac OSX server 10.5
But, we experiencing some downtime periods because of CPU overcharged.

How can we solve that problem ?

[Grumpy]

You should first login to your box via ssh and get the outputs of few commands such as... Especially during peak hours.

top -n 1
iostat -x 10 2

See sticky for more info.


If you are certain it's due to lack of CPU power, you can attempt to optimize and / or buy better hardware.
However, you should know that "lack of cpu power" doesn't necessarily translate to need better cpu as the cpu usage efficiency is highly dependent on other components.

[donpetru]

I think you need more RAM (at least 16GB) and maybe two CPU's, especially if traffic passes 1000 users (15 ... 60 minutes). I watch my VPS which 40 users online (15 minutes) already park would need more than 1Gb RAM.

For example, look this forum:
http://forum.softped...dex.php?act=idx
when I write this post has almost 2500 users online (15 minutes). There are cases when it had over 3000. And of course it is IPB 2.X. It is a forum hosted in Romania (SQL over 10Gb) and the Softpedia site has another server - from what I understand - hosted somewhere in U.S.A..

However, the idea remains that IPB 3.3.x eat a lot of resources especially if you have installed all IPB applications and they are functional (and of course, talk and a SQL database for at least a few 100Mb).
If before, in my case, could I present on a shared hosting configuration, now it seems that no VPS is not enough. I don't like that.

[SecondSight]

Hello, :)

I don't know anything about your configuration, but perhaps I can give you a piece of advice if your tables are all MyIsam...

I had some problems some time ago even though Apache and MySQL were optimized, and I had them fixed this way :
- Changing some tables (ibf_core_item_markers and ibf_core_item_markers_storage ) from MyIsam to Innodb.
- Having the ibf_sessions table placed in MEMORY.

I was told this should be done to make things work better and it worked for me. :)

[RidinHighSpeeds]

Hello, :smile:

I don't know anything about your configuration, but perhaps I can give you a piece of advice if your tables are all MyIsam...

I had some problems some time ago even though Apache and MySQL were optimized, and I had them fixed this way :

- Changing some tables (ibf_core_item_markers and ibf_core_item_markers_storage ) from MyIsam to Innodb.

- Having the ibf_sessions table placed in MEMORY.

I was told this should be done to make things work better and it worked for me. :smile:

If I had 10 members online at once, my dual-core VPS would hit a load of 2-3. After making these DB changes, the loads are around 0.5 to 0.7 with 10 members online.

Glad I found this thread!

[Grumpy]

If I had 10 members online at once, my dual-core VPS would hit a load of 2-3. After making these DB changes, the loads are around 0.5 to 0.7 with 10 members online.

Glad I found this thread!

10 members is a very low number. If changing the above DB settings made a significant improvement, it means you have a serious disk problem.

Assume each person visits 1 page per minute. That's 10 pages per minute with 10 online.
For 10 pages to cause load of 2, that's 12 seconds per page. Which is completely unacceptable.
Only by assuming they visit 12 pages per minute, an awful lot, we'd get a reasonable server load of 1 second to deliver 1 page.
To get a excellent performance, we'd have to assume as much as 120 pages per minute which is humanly impossible.

PS. This is not exact math, just estimate.

[Lauren3]

Hello, :smile:

I don't know anything about your configuration, but perhaps I can give you a piece of advice if your tables are all MyIsam...

I had some problems some time ago even though Apache and MySQL were optimized, and I had them fixed this way :

- Changing some tables (ibf_core_item_markers and ibf_core_item_markers_storage ) from MyIsam to Innodb.

- Having the ibf_sessions table placed in MEMORY.

I was told this should be done to make things work better and it worked for me. :smile:

How do you place a table in Memory ?
How to know which table may be changed to Innodb?

[Rhett]

You should really have someone look over your server for you, 700 users in 15 minutes is not much for a decent dedicated server.

As far as a limitations on IPB users, there is no limit, with a well setup single server you should expect around 2000-4000 users in 15 minutes, much more then that and (depending on specs of course), then you are looking at a separate DB server. I would not start changing table to innob at this point, first get your server setup and tuned properly.

[Grumpy]

How do you place a table in Memory ?

How to know which table may be changed to Innodb?

q1: http://community.invisionpower.com/topic/366304-speed-and-cleanup/#entry2289414
q2: Non-trivial.

[Lauren3]

q1: http://community.invisionpower.com/topic/366304-speed-and-cleanup/#entry2289414

q2: Non-trivial.

Thanks!