Slow transfer with 21140 card

[Mark Fardal]

Hi all,

I just found that UDP appears to be transmitting correctly, whereas
TCP is slow.  Thanks to Mark Hahn for the suggestion.  Now what
does this mean and what do I do?

weka# ./ttcp -s -t kaka
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  tcp  -> kaka
ttcp-t: socket
ttcp-t: connect
ttcp-t: 16777216 bytes in 36.89 real seconds = 444.09 KB/sec +++
ttcp-t: 2048 I/O calls, msec/call = 18.45, calls/sec = 55.51
ttcp-t: 0.0user 0.1sys 0:36real 0% 0i+0d 0maxrss 0+2pf 0+0csw

weka# ./ttcp -s -t kaka -u
ttcp-t: buflen=8192, nbuf=2048, align=16384/0, port=5001  udp  -> kaka
ttcp-t: socket
ttcp-t: 16777216 bytes in 1.40 real seconds = 11715.40 KB/sec +++
ttcp-t: 2054 I/O calls, msec/call = 0.70, calls/sec = 1468.72
ttcp-t: 0.0user 0.2sys 0:01real 17% 0i+0d 0maxrss 0+2pf 0+0csw

So it's tranmitting at 92 Mbits/s with udp, but only 3.5 with tcp.

> Are you sure that the hubs are full-duplex ?

> If not, you should force the card to be half-duplex, or it slows down 
> due to excessive collisions. check the docs (or module source for 
> required flags for module)

the switch is indeed full-duplex.

> However, it might well be that there
> are really cheap switches that either aren't full duplex or (more
> likely) don't have a full 100 Mbps of bandwidth per port and hence
> cannot support full duplex connections between all possible pairs of
> ports at once.  

That's an interesting theory!  I'll just shut down everyone else's
machines. :->  

To complicate matters, here is another piece of the puzzle.  There is
another machine running the 2.0.33 kernel that has 2 ethernet cards.
One is connected to the same network switch as mine (call it the
"outside network") and displays the same symptom of running only at 
< 10 Mbits/s.  The other card is connected to another, supposedly
identical switch, which is hooked up to all the machines in a single
computer lab.  That card runs at 100 Mbits/s.  The local guru suspected
a bad card, so he switched them but found nothing changed: the machine
talks to the outside network at <10, but to the local lab network at 100.

That led him to suspect the cable, which hasn't been ruled out as the
problem that machine at least.  But we ruled out the cable and/or switch
on my machine...

> We just did a check where we moved the machine over to another ethernet
> cable, where one of the properly working machines had been.  Still have
> slow transfer rates.  That might narrow down the possibilities.

thanks,
Mark Fardal
UMass