Re: Wow!!! UDB network is fast!!!! (Intel Linux off subject)

Bernd (bmeyer@cs.monash.edu.au)
Mon, 21 Oct 1996 02:48:20 +1000 (EST)

>I have experienced similarly 'slow' NFS performance on reading. It is
>possible that the OpenBSD version had cached more (or all) of the
>source file. Still, I'm skeptical of this 1MB/s transfer rate since
>that is greater the theoretical throughput limit of 10BaseT.
>
I can confirm that it seems 1M/s is not beyond the practical limit
for coax ethernet at least. But a few things are strange...

The setup:

Machine A:
A pentium 166 running linux 2.0.21; IBM DORS disk on an NCR adaptor,
and a PCI NE-2000 clone

Machine B:
An ALPHA PC164 running linux 2.1.4 at 433MHz, Seagate Baracuda disk
on Adaptec 2940 and ISA(!) NE-2000 clone, 128M of RAM

Machine B mounts a directory from machine A; For a test I timed copying
139M of MPEG layer II audio streams to machine B's local harddisk, after
previously wiping the cache with "dd if=/dev/sda of=/dev/null", which
read 1G from the disk (in 3:20 minutes, with 11% CPU load ;-).

The test took 2:09 minutes (at 9% CPU load, almost completely system
time), which means on average 1085k/s were transferred; This result
was achieved with mounting with rsize=8192,wsize=8192

However, I got the message

nfs_rpc_verify: RPC call failed: 5

for each single file. What does this mean?

Also, the reason the ALPHA has an ISA card is that it won't work with
a PCI NE-2000 clone. It would find it all right, but as soon as I did
an "ifconfig 192.32.242.21" on it, it would hang the machine hard. This
was with a variety of 2.0.x kernels; I haven't tried the PCI card on 2.1.4
yet --- does anyone know what's going on here?

Thanks,

Bernie

--
To unsubscribe: mail -s unsubscribe axp-list-request@redhat.com < /dev/null



Feedback | Store | News | Support | Product Errata | About Us | Linux Info | Search | JumpWords
No Frames | Show Frames

Copyright © 1995-1997 Red Hat Software. Legal notices