Super slow NFS
Hi,
I've been working on this problem for about week now.
I have a SMC Barricade with a K7 running Debian testing/
unstable 2.4.9 as an NFS client. The NFS server is a PPC
200MHz 604e running Debian testing/unstable 2.4.9 and is
also on the Barricade. The K7 has 100Mb Netgear311 card,
the PPC has a 10Mb old Farallon.
NFS mounts take about 15 seconds, slight lag with commands,
but extremely bad throughput. I can barely play a 192Kb OGG
file without rebuffering. Transfer rates vary wildly from
a few bytes/sec to about 300KB/sec. With scp over SSH, I
get 1.2MB/s.
I've read over the NFS-Howto and tried all the tips that seem
to apply.
Neither IPchains or IPtables are being used on either machine
so it is not a TCP fragment thing. Portsentry is running on
the client.
I have tried setting rsize/wsize=1024, 2048, and 8192 with
the same results.
>From the client kern.log:
Sep 6 02:03:09 yggdrasil kernel: nfs: server 192.168.64.115 not responding, still trying
Sep 6 02:03:09 yggdrasil kernel: nfs: server 192.168.64.115 OK
Sep 6 02:13:18 yggdrasil kernel: nfs: server 192.168.64.115 not responding, still trying
Sep 6 02:13:20 yggdrasil kernel: nfs: server 192.168.64.115 OK
Sep 6 02:13:47 yggdrasil kernel: nfs: server 192.168.64.115 not responding, still trying
Sep 6 02:13:47 yggdrasil kernel: nfs: server 192.168.64.115 OK
>From the server kern.log:
Sep 6 00:52:05 nolandia kernel: eth0: 21143 10baseT link beat good.
Sep 6 00:52:36 nolandia last message repeated 420 times
Sep 6 00:53:36 nolandia last message repeated 777 times
Sep 6 00:54:03 nolandia last message repeated 349 times
These logs look wierd, but this is the first time I've
setup NFS.
nfsstat shows the client having about 5% packet retrans.
Having NFS working would be nice, but I could settle for
the hassle of scp/SSH since the throughput is about 60X
better!
Any help would be appreciated,
--
Anthony Lau
Reply to: