I've seen this problem a lot on my 164LX machines.  It appears to be lost
interrupts on the network side controller causing retransmits of data.  I
don't have deep enough knowledge (or any time) of the 164LX issues as far
as how they differ from the other machines to be able to contribute much
else to a solution, unfortunately.  My workaround is to use 10baseT mode,
and accept the lower throughput (which is still higher than 100baseT with
the failures).

I can try and dig out more emails from my co-workers about this problem,
if it would help.

Paul

On Mon, 26 Feb 2001, Danny Chouinard, wrote:

> 
> Hi!
> 
> I'm having a funny problem on a Digital AlphaPC 164LX 533 MHz.  While
> doing network backups I'm getting awful performance.  When I'm
> transferring data both from the network and disk I'm getting about 500
> kilo-bytes per second, which is pitiful.  I'm running NetBSD 1.4T.
> 
> If I do rsh remotehost "dd if=/dev/zero bs=1024" | dd of=/dev/null
> bs=1024 I get about 6 megabytes per second, which is ok so network
> performance is OK.
> 
> If I do dd if=/dev/zero bs=1024 of=/tmp/somefile I get about 10
> megabytes per second, which is ok so disk performance is OK.
> 
> Now if I do rsh remotehost "dd if=/dev/zero bs=1024" | dd
> of=/tmp/somefile bs=1024 I get 500K-bytes/s; warm molasses speed.
> What gives?  Other dd buffer sizes (1b, 1k, 4k, 512k) do not help.
> 
> There is no load on this machine.  I get no messages logged, I've tried
> just about everything and as soon as I do network and disk I/O (either
> SCSI or IDE) it crawls.  The same tests done on a similar linux box
> give me ~5M-Bytes/s.
> 
> Please help,  Thanks.
> -- 
> Danny Chouinard <danny@cim.mcgill.ca>
> Center for Intelligent Machines, Systems Programmer, McGill U.
> 

+---------------------------------------------------+
| Paul Anderson           Public Data Queries, Inc. |
| pha@pdq.com             734-213-4964 x308         |
+---------------------------------------------------+