Re: 2.6.0 NFS-server low to 0 performance

[Mike Fedyk]

On Fri, Jan 09, 2004 at 11:08:29AM +0100, Guennadi Liakhovetski wrote:
> On Wed, 7 Jan 2004, Mike Fedyk wrote:
> 
> > Just post a few samples of the lines that differ.  Any files should be sent
> > off-list.
> 
> Ok, This is the problem:
> 
> 10:38:30.867306 0:40:f4:23:ac:91 0:50:bf:a4:59:71 ip 590: tuxscreen.grange > poirot.grange: icmp: ip reassembly time exceeded [tos 0xc0]
> 
> A similar effect was reported in 1999 with kernel 2.3.13, also between 2
> 100mbps cards. It also was occurring with UDP NFS:
> 
> http://www.ussg.iu.edu/hypermail/linux/net/9908.2/0039.html
> 
> But there were no answers, so, I am CC-ing Sam, hoping to hear, if he's
> found the reason and a cure for his problem. Apart from this message I
> didn't find any other relevant hits with Google.
> 
> Is it some physical network problem, which somehow only becomes visible
> under 2.6 now, with UDP (NFS) with 100mbps?

Find out how many packets are being dropped on your two hosts with 2.4 and
2.6.

If they're not dropping packets, maybe the ordering with a large backlog has
chagned between 2.4 and 2.6 that would keep some of the fragments from being
sent in time...
-
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/