SUMMARY: rcp speed deltas + more questions

From: Kirk Anderson (
Date: Mon Feb 03 1997 - 10:56:37 CST

Thanks to all who replied. The problem with slow file transfers ended up
being a hardware issue -- a faulty transceiver. Its replacement not only
remedied the slow transfers in one direction, it also reduced the transfer
time by a factor of nearly two (compared to the former "good" value) in
both directions.

My investigation has, however, raised more questions than it answered. If
this forum is not the appropriate place, please indicate so privately or
direct me to a "networks" forum. Our network processes real-time data, so
any significant latency introduced by the network hardware is unacceptable.

Our foremost concern is how can a transceiver operate in the degraded mode
(slow transfer time in one direction only) without failing altogether? We
have network monitoring tools (eg., HP LanProbe) and staff which failed to
detect and/or isloate the degraded transceiver. We would like to have strong
confidence in the physical layer, but that confidence is now waning.

Our second concern is how can we bench-test these devices and assure a
transfer rate within a reasonable percentage of optimum behavior? Everyone
with whom I have spoken about testing these devices knew of no commercial
(or custom) tools to verify their operation. All concluded that a two-node
test network was the best manner for testing -- it worked for me.

Another concern regards the differences between the reporting of 'netstat -i'
and 'netstat -s'. How are the input errors, out errors and collisions
tabulated for the 'netstat -i' command? I expected to able to corelate the
collisions with the retransmissions indicated from 'netstat -s' but the numbers
were never even _close_ to one another.

Finally, after replacing the bad transceiver, 'netstat -i' reported _zero_
input and output errors for that network (le0). My other nodes still reported
a small number of input errors (less than 1%). In contrast, our other network
was indicating 10% input errors. Are these a function of the transceivers
proper, or of the data, or of both? What _are_ acceptable guidelines for
input or output errors, and for collisions?

Thanks in advance,
  Kirk "full-time software engineer, part-time sysAdmin & part-time netAdmin"

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:44 CDT