[NBLUG/talk] Extremely poor network performance

[Lincoln Peters]

I have one box running Debian/testing on my network that has been 
demonstrating hideously poor performance on my Ethernet LAN.  I tried 
uploading a large directory to it via scp, and the throughput for each 
file rarely exceeds 10KB/s.  There is also a delay of several seconds 
between when one file finishes uploading and another begins.  At this 
rate, it will take several weeks to upload a 35GB directory on a 
100BaseTX LAN!


Two days ago, I tried pinging the misbehaving Linux box (60 pings total, 
at 1 ping per second) from the machine I was trying to upload from (a 
Macintosh running MacOS 10.4.5).  The results were:

round-trip min/avg/max/stdev = 0.507/285.772/911.228/233.467 ms


I had heard that this particular type of Macintosh had been reported to 
have issues with network performance, so at first I suspected it was 
having problems.  So I tried pinging two other Linux boxes from the 
Macintosh:

round-trip min/avg/max/stdev = 0.377/0.456/0.671/0.046 ms
round-trip min/avg/max/stdev = 0.451/0.624/0.934/0.060 ms

So apparently, the problem was NOT on the Mac's end.


I didn't have time to invesigate further that day, so the next day, I 
went to the misbehaving box and tried pinging two other Linux boxes on 
the LAN:

rtt min/avg/max/mdev = 0.240/548.421/1817.558/499.307 ms, pipe 2
rtt min/avg/max/mdev = 0.146/408.204/1478.301/358.105 ms, pipe 2


Then I tried pinging the misbehaving box from the Macintosh again, and 
the results were even worse:

round-trip min/avg/max/stdev = 0.433/1512.158/5762.936/1319.562 ms


As far as I could tell, there is virtually no other traffic going in or 
out of the computer that would explain these wildly varying response 
times.  But just to be sure, I disabled every server program that might 
have beeen generating network traffic.  This morning, I tried using the 
misbehaving box to ping another Linux box:

rtt min/avg/max/mdev = 0.151/4980.819/21034.876/4896.826 ms, pipe 10


Then I tried pinging my mail server (which is on the Internet, and 
therefore accessed via my router), and the results were almost shocking:

60 packets transmitted, 53 received, 11% packet loss, time 1058068ms
rtt min/avg/max/mdev = 60.530/4647.363/22086.280/4880.662 ms, pipe 13

(I indicate packet loss here because this is the only test I ran that 
experienced any packet loss.)

To make things even stranger, there was a huge lag between the 
transmission of each ping.  Although I had indicated that a total of 60 
pings should be sent out at 1-second intervals, the entire operation 
(according to the "time" command) took 18 minutes, 37.984 seconds!


To be thorough, I tried using the Macintosh to ping the same mail server:

60 packets transmitted, 60 packets received, 0% packet loss
round-trip min/avg/max/stdev = 60.243/62.168/65.857/1.105 ms


Finally, I tried restarting the network connection on this computer (as 
per "ifdown" and "ifup").  Now it seems to be fine, but I can't figure 
out why.  Seeing as how this computer is running Linux and therefore 
should not be subject to random failures, I worry that the problem will 
come up again, that and it might even get worse over time.

Any ideas what might be going on?


-- 
Lincoln Peters
<sampln at sbcglobal.net>

Of course there's no reason for it, it's just our policy.