Re: 2.6.0 NFS-server low to 0 performance

[Trond Myklebust]

På lau , 10/01/2004 klokka 17:14, skreiv Mike Fedyk:

> I have to admit, I haven't been following NFS on TCP very much.  Is the code
> in the stock 2.4 and 2.6 kernels ready for production use?  It seemed from
> what I read it was still experemental (and even marked as such in the
> config). 

The client code has been very heavily tested. It is not marked as
experimental.
The server code is marked as "officially experimental, but seems to work
well". You'll have to talk to Neil to find out what that means. In
practice, though, it performs at least as well as the UDP code.

If you are in a production environment and really don't want to trust
the TCP code, you can disable it, and use the option I mentioned earlier
of setting a low value of r/wsize.

Or better still: fix your network setup so that you don't lose all those
UDP fragments (check switches, NICs, drivers,...). The icmp time
exceeded error is a sign of a lossy network, NOT a broken NFS
implementation.

Trond
-
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/