We are running a 3.2c 2100 system.
Every now and then, our system starts crawling.
little history...Our system and everything hooked to it is
running on 10mbs, which is hooked to the main network via
a 10/100 switched port. My thought is that we just can't
get packets into the main network at 10mbs while data is
flying by at 100mbs (I'm assuming that the switched hub port)
is putting on the brakes. This slowdown is bad enough that
is stops our DecServers from talking to our Alphaserver for
over a second, then it logs everyone off (the DecServer running LAT).
I tried running tcpdump, but when the system slows down, so does
the tcpdump, when the system "freezes" so does tcpdump. I also had
tcpdump running on a linux system hooked into the same system, when
the DEC slows/freezes, the linux system doesn't, but my PC is on the
other end of the switched hub, and not connected to the same system.
I haven't tried running tcpdump from the console yet, but will since
the console isn't dependent on the network at all, just to see if
when the slowdown happens, if the console slows as well.
As well the collision light on the hub goes crazy during the slowdown,
which further makes me think it's a 10mbs vs 100mbs issue.
Unfortunatly, we get A LOT OF TRAFFIC from the 100mbs side that isn't
destined to us at all (the switch seems to be passing all packets for
a section of the building, which I thought these switches were smart
and only forwarded packets that it knew where on the other side --
either way I can't control the switches and that dept doesn't seem
to think it's a problem)
Does this sound like our 10mbs just can't get packets into the 100mbs
stream, (like if someone were pulling a large file along the main
network going 100mbs <--> 100mbs) or could this possibly be our tulip
card starting to get flaky?
I have tcpdump running 24/7 in the background (I added the -n so it
doesn't try to do a reverse DNS), as well we have tcpwrappers running
which does do a reverse DNS, I know that if the DNS isn't working
properly it can cause a slowdown to the process telneting in, but
would it also cause slowdowns to existing connections that were prior
to the slowdowns running fine, then about 3-5 minutes, the slowdown
and collisons stop, just a quickly as it started, and noone has
complained
that it takes a long time to login, just every now and then the system
seems real slow (I'm guessing it's looks slow because the system can't
get the packets to them, and telnet is a lot more forgiving in time
frame before it gives connection to host lost, as opposed to the LAT
1 sec deal).
Can the timeout for LAT disconnect where the terminal server (DS700)
panics
and logs everyone off be increased to longer than 1 second? I recall
being
told that is the maximum (and an LAT disadvantage)?
George Gallen
ggallen_at_slackinc.com
Received on Mon Oct 04 1999 - 21:44:52 NZDT