2005-01-14 23:34:26

by Jeff Blaine

Subject: NFS file xfer of size of client memory = bad scene

I have a box with 4GB. It's running 2.6.10.

I am trying to run Iozone with -g and an argument to it
that is larger than the physical memory in the NFSv3
client machine.

For instance, on this 4GB box, I am trying '-g 4224m'.
More specifically, the entire command line is:

./iozone -a -g 4224m -f /sol9box/testfile

As soon as it finally gets to trying a file size of
4194304, oom-killer steps in and starts blasting
processes off my machine (WTF? Great.)

Brings up 'oom-killer' instances under 2.6.10 and the
whole benchmark run dies, sometimes along with my
shell, my X session, ANY net services at all (can't
SSH into the box... can't get console to wake up...),
etc.

Exact same command line worked in Linux 2.4.21 loaded
on this same exact machine with the same NFSv3 server
being talked to.



-------------------------------------------------------
The SF.Net email is sponsored by: Beat the post-holiday blues
Get a FREE limited edition SourceForge.net t-shirt from ThinkGeek.
It's fun and FREE -- well, almost....http://www.thinkgeek.com/sfshirt
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs

2005-01-15 05:43:22

by Trond Myklebust

Subject: Re: NFS file xfer of size of client memory = bad scene

fr den 14.01.2005 Klokka 18:34 (-0500) skreiv Jeff Blaine:
> I have a box with 4GB. It's running 2.6.10.
>
> I am trying to run Iozone with -g and an argument to it
> that is larger than the physical memory in the NFSv3
> client machine.
>
> For instance, on this 4GB box, I am trying '-g 4224m'.
> More specifically, the entire command line is:
>
> ./iozone -a -g 4224m -f /sol9box/testfile
>
> As soon as it finally gets to trying a file size of
> 4194304, oom-killer steps in and starts blasting
> processes off my machine (WTF? Great.)
>

I suggest you take this bug up to LKML or the memory management lists.
It is the responsibility of the VM to initiate flushing of data when the
memory pressure gets too high: the NFS subsystems have little control
over this matter.

Cheers,
Trond

--
Trond Myklebust <[email protected]>



-------------------------------------------------------
The SF.Net email is sponsored by: Beat the post-holiday blues
Get a FREE limited edition SourceForge.net t-shirt from ThinkGeek.
It's fun and FREE -- well, almost....http://www.thinkgeek.com/sfshirt
_______________________________________________
NFS maillist - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs