Use 'ifstat -z' to zero the stats and see if the errors are still currently occurring. If they are occurring, and if still CRC's, check hardware starting with the easiest first: cable.
This would be why UDP *seems* to work while TCP "blows the whistle" on the errors.
----- Original Message ----- From: James Beal james.beal@sanger.ac.uk To: Webster, Stetson Cc: dh3@sanger.ac.uk dh3@sanger.ac.uk; toasters@mathworks.com toasters@mathworks.com; Blakemore, Steven Sent: Wed Jun 11 13:40:52 2008 Subject: Re: strange behaviour, Linux and NFS on NTFS qtree
Webster, Stetson wrote:
Any errors or strange numbers in 'ifstat -av' ?
netapp1a*> ifstat -v e4a
-- interface e4a (7 days, 10 hours, 23 minutes, 16 seconds) --
RECEIVE Frames/second: 311 | Bytes/second: 2251k | Errors/minute: 4864 Discards/minute: 0 | Total frames: 277m | Total bytes: 63451m Total errors: 36939k | Total discards: 0 | Multi/broadcast: 108k No buffers: 0 | Non-primary u/c: 101k | Tag drop: 0 Vlan tag drop: 0 | Vlan untag drop: 0 | Mac octets: 132g UCast pkts: 1187m | MCast pkts: 17000 | BCast pkts: 91533 CRC errors: 36939k | Bus overrun: 0 | Alignment errors: 0 Long frames: 0 | Jabber: 0 | Pause frames: 0 Runt frames: 0 | Symbol errors: 0 | Jumbo frames: 0 TRANSMIT Frames/second: 433 | Bytes/second: 1506k | Errors/minute: 0 Discards/minute: 0 | Total frames: 563m | Total bytes: 3323g Total errors: 0 | Total discards: 0 | Multi/broadcast: 7559 Queue overflows: 0 | No buffers: 0 | Frames queued: 0 Buffer coalesces: 4 | MTUs too big: 0 | Mac octets: 3462g UCast pkts: 2452m | MCast pkts: 2922 | BCast pkts: 4637 Bus underruns: 0 | Pause frames: 0 | Jumbo frames: 0 LINK_INFO Current state: up | Up to downs: 1 | Speed: 10000m Duplex: full | Flowcontrol: full
It's using a 10Gig card with no vif's ( I believe ) . When we stress the system we see the following on the console , although when we do this test we don't see the problem ( We see it when I throw about 1000 cores at the system and the 10Gig card is at about 60% and the CPU is around 100% ). I have asked our presales engineer about this.
XXX restart_tx restart_offloadq XXX restart_tx XXX restart_tx restart_offloadq XXX restart_tx XXX restart_tx
----- Original Message ----- From: Dave Holland dh3@sanger.ac.uk To: Webster, Stetson Cc: toasters@mathworks.com toasters@mathworks.com Sent: Wed Jun 11 11:58:38 2008 Subject: Re: strange behaviour, Linux and NFS on NTFS qtree
On Wed, Jun 11, 2008 at 11:30:22AM -0400, Webster, Stetson wrote:
What ONTAP release, what Linux kernel, what NFS mount options?
I knew I'd missed things...
It's ONTAP 7.2.4. I can upgrade to 7.2.5 if that'll help.
The Linux kernel is 2.6.18-6-686 (Debian 4.0, 2.6.18.dfsg.1-18etch4), and the problem also shows with 2.6.8-2-686-smp and 2.6.5--286tg3susesfs.
I noticed this when mounting with proto=tcp,vers=3,rsize=8192,wsize=8192. I'd also tried UDP, and the problem persisted.
But after your email I tried vers=2 and the problem goes away (with both TCP and UDP) which is interesting indeed. Although with the crazy size files and filesystems around here, NFSv3 is very desirable.