Re: Fragment flooding in 2.4.x/2.5.x

Trond Myklebust (trond.myklebust@fys.uio.no)
Mon, 1 Jul 2002 14:14:50 +0200


On Friday 28 June 2002 20:21, Alexey Kuznetsov wrote:
> Hello!
>
> > suddenly jump to ~4.5MB/s (peak was 5MB/s).
>
> Hmm.. it is funny that you were satisfied with previous value
> and it is funny that it still does not saturate link.

It is only recently that we have come far enough with implementing things like
round trip timing, congestion control, etc. to really start noticing these
effects. Without the RTT scheme on the UDP link, you simply don't see it (you
only notice a large difference with TCP).

Note that this covers the discrepancy between NFS over TCP and NFS over UDP
against that particular machine, so I do not expect further improvements.
The main reason why I don't expect to saturate the link is that these are NFS
*writes*, hence random things like file semaphore contentions, disk access
and write speeds etc. on the server, pop up.

> Of course. If you noticed this year or two or three ago, it would be even
> an urgent problem. But until now it was problem with status of "well-known
> bogosity which requires some sane solution but can wait for some good idea
> for infinite time because of absence of any real applications sensing it"
> :-)

I've now got the application and a demonstration of what kind of fix is
needed. I hope you and Dave can work out a better patch in for 2.4.20-pre
;-)...

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