OpenVZ Forum: Support » system limits

Home » General » Support » system limits

Show: Today's Messages :: Show Polls :: Message Navigator
E-mail to friend

system limits [message #3070]

Thu, 11 May 2006 01:47

John Kelly
Messages: 97
Registered: May 2006
Location: Palmetto State

Member

For my needs, openvz appears to offer a better solution than xen. But I worry about system limits:

http://www.redhat.com/archives/rhl-devel-list/2006-April/msg 01356.html

With a large number of threads/pids, is it true? Can I exhaust 32 bit virtual address space?

Report message to a moderator

Re: system limits [message #3074 is a reply to message #3070]

Thu, 11 May 2006 08:27

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

the link you provided tells about per-process virtual address space limitations. It has nothing to do with OpenVZ, but rather with i686 arch. So what is your exact question, what do you mean? Whether this limitation will be gloabal-wide and effect OpenVZ? The answer is NO then.

http://static.openvz.org/userbars/openvz-developer.png

Report message to a moderator

Re: system limits [message #3079 is a reply to message #3074]

Thu, 11 May 2006 10:35

John Kelly
Messages: 97
Registered: May 2006
Location: Palmetto State

Member

dev wrote on Thu, 11 May 2006 04:27

OK, thanks for clarification.

To use openvz for mass virutal hosting, with large number of VPS and PIDs, what is the practical upper limit for maximum number of PIDs, before scheduling overhead becomes a problem?

Report message to a moderator

Re: system limits [message #3081 is a reply to message #3079]

Thu, 11 May 2006 10:38

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

PIDs itself do no introduce scheduling overhead.
on scheduling only number of _running_ processes influences.

Report message to a moderator

Re: system limits [message #3082 is a reply to message #3081]

Thu, 11 May 2006 10:48

John Kelly
Messages: 97
Registered: May 2006
Location: Palmetto State

Member

dev wrote on Thu, 11 May 2006 06:38

PIDs itself do no introduce scheduling overhead.
on scheduling only number of _running_ processes influences.

I admit I don't know much about linux kernel. But suppose I have a 40 to 1 ratio, total PIDs vs. running PIDs. And suppose I have 20,000 total PIDs, 500 running PIDs.

Do you think linux and the CFQ scheduler can handle that?

Report message to a moderator

Re: system limits [message #3090 is a reply to message #3082]

Thu, 11 May 2006 20:59

RapidVPS
Messages: 32
Registered: January 2006

Member

Hi, I really doubt you will support 20,000 pids or 500 running pids on normal server hardware (assuming you meant those numbers literally)- for example on a dual core dual opteron w/ 8GB mem. From my experience, you can have 4000-6000 total processes in such a production system and it will still be handled OK, that is probably 10-25 running procs average. 500 running processes would kill any system.

To answer your question, I have pushed vz to 10,000 pids and ~50 running average in a test case and the server will be slow but alive. All you really need to be concerned about is the number of running processes, I dont think total pids plays a factor in the 2.6 scheduler, they did in 2.4.

[Updated on: Fri, 12 May 2006 00:25]

Report message to a moderator

Re: system limits [message #3098 is a reply to message #3082]

Fri, 12 May 2006 08:18

dev
Messages: 1693
Registered: September 2005
Location: Moscow

Senior Member

CFQ scheduler is about disk I/O, not pids and CPU.
RapidVPS provided you a correct and good info about running processes. If you have little running processes than even 200,000 of pids is not a problem.

Report message to a moderator

Previous Topic:	Host firewall -- SOLVED
Next Topic:	chkpnt does not work

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Fri Jul 12 13:23:44 GMT 2024

Total time taken to generate the page: 0.02359 seconds