...
3c900.c:v0.40 4/16/97 becker@cesdis.gsfc.nasa.gov
loading device 'eth0'...
eth0: 3Com 3c905 Boomerang 100baseTx at 0xf480, 00:60:97:c0:81:90, IRQ 11
8K word-wide RAM 3:5 Rx:Tx split, autoselect/MII interface.
Rx Pacing bug exists, disabling bus-master receives.
eth0: Overriding PCI latency timer (CFLT) setting of 64, new value is 248.
eth0 UP fl=00001043 pa=06B8E380/0000FFFF brd=FFFFE380 dst=00000000
eth0 changes netmask 0000FFFF -> 00FFFFFF
eth0 changes brd FFFFE380 -> FFB8E380
route forgot to specify route netmask.
portmap: RPC call returned error 111
RPC: task of released request still queued!
RPC: (task is on xprt_pending)
portmap: RPC call returned error 111
RPC: task of released request still queued!
RPC: (task is on xprt_pending)
portmap: RPC call returned error 111
RPC: task of released request still queued!
RPC: (task is on xprt_pending)
Installing knfsd (copyright (C) 1996 okir@monad.swb.de).
nfs warning: mount version older than kernel
nfs warning: mount version older than kernel
nfs warning: mount version older than kernel
...
The NFS warnings are from amd (does anyone know what I need to do in order
to update amd to use the new kernel NFS properly?). I assume (perhaps
incorrectly) that the RPC errors are related to why there is no
/proc/net/rpc directory.
Also, NFS writes are still 5-10 times slower than what I'd hoped to
see. Even when writing from a Pentium Pro to an Alpha (across 10 Mbps
ethernet), I get 107 KBps, but when doing the same thing from an ancient
Sparc 2 to the Alpha I get 349 KBps! NFS reads are about 800 KBps for
both the PPro and the Sparc 2, and using "rcp" to copy from the PPro to
the Alpha gives about 570 KBps. These figures are for non-amd amounts
(amd doesn't seem to harm the performance, though), using rsize and wsize
of 8192. I'm also using a highly hacked 3c59x.c driver for my 3c905 card
(got sick of rebooting every few hours from transmitter access conflicts)
on the PPro, but I doubt that is causing the performance problems.
I also notice that writing to a Sun (either SunOS or Solaris), will result
in some very high load averages/iowait statistics. The Alpha doesn't do
that, but it still doesn't give much better performance figures either.
Does anyone have any ideas about what I might be doing wrong or what I
could try?
Many thanks,
Brent Nelson
Systems Manager
Astronomy Department
University of Florida