Mysterious slow NFS access

[trifo]

Hi Folks,

I have a set of four AIX hosts acting as NFS clients and a Linux NFS server. There is three of the four AIX hosts which can write to the Linux NFS share with a reasonable speed (60-70MB/s), BUT the remaining one AIX host is terribly slow (~3.8MB/s). The problem is that the slow one should write a very huge amount of data on a daily basis (Oracle backups, more than 2TB)

I have already checked everything I was able to. There were no difference in
- nfso
- vmo
- sys0 attributes
- name resolution (netsvc.conf an hosts files are identical)
- network interface config
- routing config (are defaults)

The main problem is that the network throughput speeds are correct in any directions using any OTHER protocol. Tried using FTP and RSH, transmitting from hostA:/dev/zero to hostB:/dev/null

The slow host - let's call is ProdDB - shows the same slow speed acting as NFS client and as NFS server as well.

Tried to watch nfsstat outputs, but there were no retransmissions, nor any other signs of failing. There were - almost - write requests only.

Do you have any idea on what to check, what to set or where to buy a rope to hang myself?

Thangs in advance,

--Trifo

 

[amoser]

Trifo,
Have you checked the port settings on the network switch. In my experience, autonegotiate does not work well. I set both the NIC interface and switch ports to correct speed and duplex.