On Feb 11, 2007  23:51 -0800, Sridharan Ramaswamy (srramasw) wrote:
> Thanks Andreas. I'll try a lower value for the num_threads=NNN. How low
> can these values be?

On a very small UML system there are only 2 threads per service. 

> > -----Original Message-----
> > From: Andreas Dilger [mailto:[EMAIL PROTECTED] 
> > Sent: Sunday, February 11, 2007 11:39 PM
> > To: Sridharan Ramaswamy (srramasw)
> > Cc: [email protected]
> > Subject: Re: [Lustre-discuss] Number of processes in Lustre tunable?
> > 
> > On Jan 23, 2007  10:39 -0800, Sridharan Ramaswamy (srramasw) wrote:
> > > I'm tring to measure the overhead of having Lustre on the 
> > system.  Among
> > > other things, I noticed it starts lot of processes,
> > >  
> > > root      4565     1  0 Jan03 ?        00:00:00 [ll_mdt_attr_14]
> > > root      4636     1  0 Jan03 ?        00:00:01 [ll_ost_13]
> > > root      4651     1  0 Jan03 ?        00:00:08 [ll_ost_io_13]
> > > root      4550     1  0 Jan03 ?        00:00:03 [ll_mdt_14]
> > > root      4516     1  0 Jan03 ?        00:00:00 [ldlm_cb_31]
> > > 
> > > I can account about 200 processes related to Lustre on an 
> > OST node. That
> > > seems a lot considering we would like to run other 
> > applications on the
> > > same node. Is it possible to tune these number of processes 
> > down? Say
> > > reduce 32 ldlm process to about 8. What kind of performance 
> > degradation
> > > to expect if we can do that?
> > 
> > You can reduce the MDS and OST process count via modprobe.conf:
> > 
> > option mds mds_num_threads=NNN
> > option ost ost_num_threads=NNN
> > 
> > If this isn't good enough, you can change the count at compile time
> > in lustre/include/linux/lustre-net.h
> > 
> > In 1.6 the number of threads will be more dynamic, depending on load.
> > 
> > Cheers, Andreas
> > --
> > Andreas Dilger
> > Principal Software Engineer
> > Cluster File Systems, Inc.
> > 

Cheers, Andreas
--
Andreas Dilger
Principal Software Engineer
Cluster File Systems, Inc.

_______________________________________________
Lustre-discuss mailing list
[email protected]
https://mail.clusterfs.com/mailman/listinfo/lustre-discuss

Reply via email to