Per process accounting for nfs

Per process accounting for nfs

[Sean Brisbane]
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==

Re: Per process accounting for nfs

[Benjamin Coddington]

[-- Attachment #1: Type: text/plain, Size: 767 bytes --]

On Thu, 2 Apr 2015, Sean Brisbane wrote:

> Dear All,
>
> I'm currently trying to identify a mechanism to measure NFS I/o per process on a multi user NFS client machine. Lots of googling tells me this is not straight forward. Installing a kernel or module is an option if necessary, though we are also supporting afs and lustre on the same kernel (RHEL 6 based). The purpose of this is to gather accurate metrics on the workload types we run to feed into what hardware we need to buy. Can anyone give me any pointers as to how to proceed?
>
> Thanks,
> SeanN?????r??y??????X??ǧv???)޺{.n??????{???????n?r??????????????&????G?????\x03(?階?ݢj"????^[m????????ޖ???f???h???~?m?

Hi Sean,

Systemtap can do that.. and ftrace might also work for that.  Check those
out.

Ben

RE: Per process accounting for nfs

[Sean Brisbane]

Dear Ben,=0A=
=0A=
System tap is exactly what I was looking for, and more.  Ftrace looks promi=
sing too, but systemtap has an example script that nearly matches the use c=
ase.=0A=
=0A=
Thank you for your help.=0A=
Sean=0A=
________________________________________=0A=
From: Benjamin Coddington [bcodding@redhat.com]=0A=
Sent: 02 April 2015 20:50=0A=
To: Sean Brisbane=0A=
Cc: linux-nfs@vger.kernel.org=0A=
Subject: Re: Per process accounting for nfs=0A=
=0A=
On Thu, 2 Apr 2015, Sean Brisbane wrote:=0A=
=0A=
> Dear All,=0A=
>=0A=
> I'm currently trying to identify a mechanism to measure NFS I/o per proce=
ss on a multi user NFS client machine. Lots of googling tells me this is no=
t straight forward. Installing a kernel or module is an option if necessary=
, though we are also supporting afs and lustre on the same kernel (RHEL 6 b=
ased). The purpose of this is to gather accurate metrics on the workload ty=
pes we run to feed into what hardware we need to buy. Can anyone give me an=
y pointers as to how to proceed?=0A=
>=0A=
> Thanks,=0A=
=0A=
=0A=
Hi Sean,=0A=
=0A=
Systemtap can do that.. and ftrace might also work for that.  Check those=
=0A=
out.=0A=
=0A=
Ben=0A=