Mailing List Archive: Interchange: users

StartServers and RAM

Index | Next | Previous | View Threaded

emailgrant at gmail

Jun 14, 2013, 1:21 AM

Post #1 of 5 (87 views)
Permalink

StartServers and RAM

I'm experimenting with different values for StartServers and I've
discovered that most of the resident memory associated with IC
processes seems to be shared. 'top' shows each of my processes have
over 100M resident memory and yet doubling or tripling the
StartServers value barely changes the -/+ buffers/cache values in
'free'. I'm also using MaxRequestsPerChild 100 and ChildLife 3600.

Has anyone found a good method/formula for maximizing StartServers and
preventing OOM? How much RAM should you have for StartServers 150 for
example?

Is RAM the only consideration when tuning StartServers?

- Grant

_______________________________________________
interchange-users mailing list
interchange-users [at] icdevgroup
http://www.icdevgroup.org/mailman/listinfo/interchange-users

jon at endpoint

Jun 14, 2013, 2:21 AM

Post #2 of 5 (74 views)
Permalink

Re: StartServers and RAM [In reply to]

On Fri, 14 Jun 2013, Grant wrote:

> I'm experimenting with different values for StartServers and I've
> discovered that most of the resident memory associated with IC processes
> seems to be shared. 'top' shows each of my processes have over 100M
> resident memory and yet doubling or tripling the StartServers value
> barely changes the -/+ buffers/cache values in 'free'. I'm also using
> MaxRequestsPerChild 100 and ChildLife 3600.
>
> Has anyone found a good method/formula for maximizing StartServers and
> preventing OOM? How much RAM should you have for StartServers 150 for
> example?
>
> Is RAM the only consideration when tuning StartServers?

RAM and CPU and workload (I/O or CPU bound?).

A good rule of thumb would be to have maybe 2-5 Interchange child
processes per CPU core. It's hard to see the use of more than that unless
you have a heavily database-bound setup where many IC children need to be
serving a request but wait around a lot not burning CPU. But if you're
CPU-bound in Interchange and things get busy, they'll just be fighting
against each other if you get too many at the same time.

Jon

--
Jon Jensen
End Point Corporation
http://www.endpoint.com/

_______________________________________________
interchange-users mailing list
interchange-users [at] icdevgroup
http://www.icdevgroup.org/mailman/listinfo/interchange-users

emailgrant at gmail

Jun 14, 2013, 2:55 AM

Post #3 of 5 (76 views)
Permalink

Re: StartServers and RAM [In reply to]

>> I'm experimenting with different values for StartServers and I've
>> discovered that most of the resident memory associated with IC processes
>> seems to be shared. 'top' shows each of my processes have over 100M
>> resident memory and yet doubling or tripling the StartServers value barely
>> changes the -/+ buffers/cache values in 'free'. I'm also using
>> MaxRequestsPerChild 100 and ChildLife 3600.
>>
>> Has anyone found a good method/formula for maximizing StartServers and
>> preventing OOM? How much RAM should you have for StartServers 150 for
>> example?
>>
>> Is RAM the only consideration when tuning StartServers?
>
>
> RAM and CPU and workload (I/O or CPU bound?).
>
> A good rule of thumb would be to have maybe 2-5 Interchange child processes
> per CPU core. It's hard to see the use of more than that unless you have a
> heavily database-bound setup where many IC children need to be serving a
> request but wait around a lot not burning CPU. But if you're CPU-bound in
> Interchange and things get busy, they'll just be fighting against each other
> if you get too many at the same time.

Understood, thank you Jon. Is there a good method or monitoring
program for figuring out the bottleneck?

- Grant

_______________________________________________
interchange-users mailing list
interchange-users [at] icdevgroup
http://www.icdevgroup.org/mailman/listinfo/interchange-users

jon at endpoint

Jun 14, 2013, 3:40 AM

Post #4 of 5 (76 views)
Permalink

Re: StartServers and RAM [In reply to]

On Fri, 14 Jun 2013, Grant wrote:

> Is there a good method or monitoring program for figuring out the
> bottleneck?

I'd start by watching top to see whether processes are CPU or I/O bound.
Induce high load with ab, siege, etc. But remember that different pages
may have different performance characteristics depending on what they do:
lots of complex ITL, database stuff, remote web service calls, etc.

Jon

--
Jon Jensen
End Point Corporation
http://www.endpoint.com/

_______________________________________________
interchange-users mailing list
interchange-users [at] icdevgroup
http://www.icdevgroup.org/mailman/listinfo/interchange-users

emailgrant at gmail

Jul 18, 2013, 11:29 AM

Post #5 of 5 (36 views)
Permalink

Re: StartServers and RAM [In reply to]

>> Is there a good method or monitoring program for figuring out the
>> bottleneck?
>
>
> I'd start by watching top to see whether processes are CPU or I/O bound.
> Induce high load with ab, siege, etc. But remember that different pages may
> have different performance characteristics depending on what they do: lots
> of complex ITL, database stuff, remote web service calls, etc.

I've been experimenting with munin too and it is nice to have the insight.

- Grant

_______________________________________________
interchange-users mailing list
interchange-users [at] icdevgroup
http://www.icdevgroup.org/mailman/listinfo/interchange-users

Index | Next | Previous | View Threaded

Interested in having your list archived? Contact Gossamer Threads