Re: bdflush/rpciod high CPU utilization, profile does not make sense

From: Jakob Oestergaard
Date: Thu Apr 07 2005 - 10:29:40 EST


On Wed, Apr 06, 2005 at 05:28:56PM -0400, Trond Myklebust wrote:
...
> A look at "nfsstat" might help, as might "netstat -s".
>
> In particular, I suggest looking at the "retrans" counter in nfsstat.

When doing a 'cp largefile1 largefile2' on the client, I see approx. 10
retransmissions per second in nfsstat.

I don't really know if this is a lot...

I also see packets dropped in ifconfig - approx. 10 per second... I
wonder if these two are related.

Client has an intel e1000 card - I just set the RX ring buffer to the
max. of 4096 (up from the default of 256), but this doesn't seem to help
a lot (I see the 10 drops/sec with the large RX buffer).

I use NAPI - is there anything else I can do to make the card not drop
packets? I'm just assuming that this might at least be a part of the
problem, but with large RX ring and NAPI I don't know how much else I
can do to not make the box drop incoming data...

> When you say that TCP did not help, please note that if retrans is high,
> then using TCP with a large value for timeo (for instance -otimeo=600)
> is a good idea. It is IMHO a bug for the "mount" program to be setting
> default timeout values of less than 30 seconds when using TCP.

I can try that.

Thanks!

--

/ jakob

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/