Based on a suggestion from one of the list members, I ran some more tests
to help narrow down the problem:
I ran some tests within the same building to remove the possibility of
selective throttling as the data winds it's way through the net
on the system of interest (i.e. sftp localhost), I get like 10Mb/s, so the
server doesn't seem pooched
the computer is actually the head node of a small cluster and I get the
results:
node2 to to master(gutowsky) via sftp: starts at 200 KB/s and ends on a
cumulative rate of 100 KB/s, which is acceptable I suppose
node2 to master(gutowsky) via iperf: 4.49 Mbits/sec (560 KB/s ?)
so the inter-node network doesn't quite perform as I would expect of a 100
Mb network, but since I didn't boot the users to run the test, it's
probably okay (and good enough, the calc.'s aren't transfer-intensive)
from the master-node(gutosky) to another comp. in the building (either
direction), I get via sftp an initial transfer rate of 100KB/s,
with the same two computers, but this quickly drops off and it keeps
stalling for 5 seconds at a time and then up to 10 Kb/s, then stall, etc.
Between these two, I get a rate with iperf of 6.8 Mbits/s (850 Kb/s ?)
the relative data of iperf/sftp seems to point to an sftp problem, unless
selective throttling of ssh traffic is an issue (which I don't think is
true within the chemistry building)
The network performance isn't what it should be regardless, so maybe if I
fix that, the whole issue will be solved, but I don't have time to try a
newer kernel now....
(p.s., tried the usual net-problem charachters of switch, full/half duplex
mode, and a different kernel module (but still 2.4 kernel))
any pointers in the right direction would be appreciated,
KH
kjharris@sdf.lonestar.org
SDF Public Access UNIX System - http://sdf.lonestar.org
