Linksys Lite-On and 0.90 vs. 0.90f (patched) vs. 0.90p

Eric Ding ericding@MIT.EDU
Fri Feb 19 13:18:44 1999


Hi all,

I've got several of the Linksys 82c169 NIC's on my intranet, and have
observed some inexplicably (at least by me) poor performance.

Machine 10.0.0.1 is a 486DX4-100 that serves as our router to our cable
  modem network, 4 MB RAM, running the 0.90f patched driver.
Machine 10.0.0.3 is a 5x86-133 on a VIP motherboard, 32 MB RAM.

When I run netperf from 10.0.0.1 to 10.0.0.3, I get these results:

TCP STREAM TEST to 10.0.0.3
Recv   Send    Send                          
Socket Socket  Message  Elapsed              
Size   Size    Size     Time     Throughput  
bytes  bytes   bytes    secs.    10^6bits/sec  

 65535  65535  65535    10.39       17.78

I *think* this is really low because of the motherboard and such, as doing
a test to 10.0.0.1 gives similar numbers.  Is this a good assumption?

The really bad performance comes when I run netperf to go from 10.0.0.3
(with the 0.90 driver) back to 10.0.0.1:

TCP STREAM TEST to 10.0.0.1
Recv   Send    Send                          
Socket Socket  Message  Elapsed              
Size   Size    Size     Time     Throughput  
bytes  bytes   bytes    secs.    10^6bits/sec  

 65535  65535  65535    10.39       1.96   

You think that's bad?  If I upgrade to the patched 0.90f or the 0.90p
driver, I get absolutely abominable results:

 65535  65535  65535    11.05       0.24   

By the way, pinging to 10.0.0.3 from itself yields the following:

 65535  65535  65535    10.00      80.47

Can anyone help me figure out what's going on? 

Thanks,
Eric