[vortex] RX overrun with 3COM 3c982

[Donald Becker]

On Wed, 13 Nov 2002, Claude Pignol wrote:
> Donald Becker wrote:
> >On Tue, 12 Nov 2002, Claude Pignol wrote:

> >>I have loaded the new driver on all the nodes.
> >>The performance using PVFS for example are correct but I still have some 
> >>overrun on the nodes
> >>that have the more network load (io nodes)
> >>Here are the stat for the io nodes:
> >>
> >>RX packets:4920572 errors:0 dropped:0 overruns:2040 frame:0
> >>TX packets:4444957 errors:0 dropped:0 overruns:0 carrier:0
> >
> >What is the detection message?

> 3c59x.c:v0.99X 6/21/2002 Donald Becker, becker@scyld.com
> http://www.scyld.com/network/vortex.html
> eth0: 3Com 3c982 Server Tornado at 0x2400, 00:e0:81:22:b0:6a, IRQ 18
> 8K buffer 5:3 Rx:Tx split, autoselect/Autonegotiate interface.
> MII transceiver found at address 24, status 782d.
> Using bus-master transmits and whole-frame receives.

OK, there is nothing unusual in the configuration.
Is this a single or dual NIC system?  (The '982 ID is often used on dual
NIC systems).

> >What does /proc/net/dev report?
> 
> > face |bytes    packets errs drop fifo frame...
> >  eth0:79819430 5177203    0    0 2040     0   

OK, this is the useful output.

The only source for FIFO errors with this speicific chip is the errors
reported in Window 6 offset 5.
These errors can occur either because
   The chip ran out of PCI bandwidth
   The driver ran out of receive buffers, causing the PCI Rx transfers
   (the "upload engine") to stall.

Are you running another high-bandwidth PCI device on the system?  Video
cards are the usual offenders.

Have you tried increasing the kernels free memory?

2.2 kernels
   echo 500 1000 2000 > /proc/sys/vm/freepages
and 2.4 kernels
   echo "100 500 200" > /proc/sys/vm/bdflush


-- 
Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Scyld Beowulf cluster system
Annapolis MD 21403			410-990-9993