Low NFS transfer speed

[m***@gmail.com]

Hi folks,

I have an annoying problem with NFS. Here's the network setup:

Linux NFS server <--- 100 Mbps link ---> NetBSD NFS client

I have this configuration:
- Server is running rpcbind, rpc.nfsd, rpc.lockd, rpc.statd,
rpc.mountd, rpc.idmapd, rpc.rquotad.
/etc/exports:
/export/nfs 192.168.1.0/255.255.255.0(rw,insecure,sync)

- Client is running rpcbind, rpc.lockd, rpc.statd.
/etc/rc.conf:
lockd=YES
statd=YES
rpcbind=YES
nfs_client=YES
/etc/fstab:
linux:/export/nfs /mnt/nfs nfs rw,tcp,intr,rsize=32768,wsize=32768 0 0

Things work, but all transfers are incredibly slow. I ran tcpdump on the
server side and I got this:

---snip---

16:02:56.053674 IP netbsd.localnet.site.1017 > linux.localnet.site.nfs: . ack
1337277 win 48836
16:02:56.053687 IP linux.localnet.site.nfs > netbsd.localnet.site.1936942451:
reply Unknown rpc response code=544499809 1460
16:02:56.053692 IP linux.localnet.site.nfs > netbsd.localnet.site.1952805478:
reply Unknown rpc response code=1635087471 1460
16:02:56.053697 IP linux.localnet.site.nfs > netbsd.localnet.site.1029448809:
reply Unknown rpc response code=1718183775 1460
16:02:56.053703 IP linux.localnet.site.nfs > netbsd.localnet.site.1650549611:
reply Unknown rpc response code=1852059146 1460
16:02:56.053825 IP linux.localnet.site.nfs > netbsd.localnet.site.543518069:
reply Unknown rpc response code=157708142 1460
16:02:56.053948 IP linux.localnet.site.nfs > netbsd.localnet.site.1835285794:
reply Unknown rpc response code=540946276 1460
16:02:56.061437 IP netbsd.localnet.site.1017 > linux.localnet.site.nfs: . ack
1346037 win 40076
16:02:56.061447 IP linux.localnet.site.nfs > netbsd.localnet.site.151587106:
reply Unknown rpc response code=1229866810 1460
16:02:56.061453 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460
16:02:56.061459 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460
16:02:56.061464 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460
16:02:56.061586 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply
Unknown rpc response code=7697268 1460
16:02:56.061708 IP linux.localnet.site.nfs > netbsd.localnet.site.1313293644:
reply Unknown rpc response code=572801864 1460
16:02:56.061832 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460
16:02:56.078249 IP netbsd.localnet.site.1017 > linux.localnet.site.nfs: . ack
1356257 win 62752
16:02:56.078260 IP linux.localnet.site.nfs > netbsd.localnet.site.543257376:
reply Unknown rpc response code=1869374752 1460
16:02:56.078266 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460
16:02:56.078271 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply
Unknown rpc response code=7697268 1460
16:02:56.078276 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460
16:02:56.078400 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460
16:02:56.078522 IP linux.localnet.site.nfs > netbsd.localnet.site.1430869024:
reply Unknown rpc response code=1919251573 1460
16:02:56.078646 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460
16:02:56.078768 IP linux.localnet.site.nfs > netbsd.localnet.site.0: reply ok
1460

---snip---

I'm not sure I understand tcpdump output well. Does this actually mean that
the NFS server is trying to send data to ports beyond 65535? And if yes, do
you have any idea why? Do you think tweaking timeo might help?

Best regards.

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[Antti Kantee]

Post by m***@gmail.com
Hi folks,
Linux NFS server <--- 100 Mbps link ---> NetBSD NFS client
- Server is running rpcbind, rpc.nfsd, rpc.lockd, rpc.statd,
rpc.mountd, rpc.idmapd, rpc.rquotad.
/export/nfs 192.168.1.0/255.255.255.0(rw,insecure,sync)
- Client is running rpcbind, rpc.lockd, rpc.statd.
lockd=YES
statd=YES
rpcbind=YES
nfs_client=YES
linux:/export/nfs /mnt/nfs nfs rw,tcp,intr,rsize=32768,wsize=32768 0 0
Things work, but all transfers are incredibly slow. I ran tcpdump on the

What is "incredibly slow"? What version of NetBSD?

(general question because i'm too lazy to look at the code at the moment:
does rsize and wsize make a difference on tcp mounts?)

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de

[David Laight]

Post by m***@gmail.com
I'm not sure I understand tcpdump output well. Does this actually mean that
the NFS server is trying to send data to ports beyond 65535? And if yes, do
you have any idea why? Do you think tweaking timeo might help?

No, tcpdump puts an RPC? number in that field.

David

[m***@gmail.com]

Post by Antti Kantee
What is "incredibly slow"? What version of NetBSD?
does rsize and wsize make a difference on tcp mounts?)

"Incredibly slow" means I download pkgsrc-current.tgz approximately five times
as fast from ftp.netbsd.org (~30Kbps) than from the local NFS server. This
occurs on both NetBSD 2.1 and NetBSD 3.1. As to rsize/wsize, bigger values
give better results. I haven't tried using NFS over UDP, so I don't know
whether it's faster than TCP transport or not. Googling for "Unknown rpc
response code" doesn't help much.

Best regards.

--
Posted automagically by a mail2news gateway at muc.de e.V.
Please direct questions, flames, donations, etc. to news-***@muc.de