Strange network problem, PLEASE HELP!
- From: "AnthonyM" <anthony.altemara@xxxxxxxxx>
- Date: 27 Jun 2006 20:02:19 -0700
I am having a strange network problem on our server testing network!
We run CIFS sessions from each client to each server under load for at
least
16 hours at a time. We may have 150 servers being tested at any given
time.
We have a simple network of switches in a hub/spoke configuration, no
redundant links or circular paths. We have 1 "core" switch, an extreme
summit 5i, and 4 other summit 5i's connected to the 'core' switch. They
all
have the latest firmware and default settings to make them 'dumb'
switches.
Connected to each of the summits are about 6 links, each going to a
separate
rack of servers. Each rack has a dumb switch (netgear / etc..) which
all the
servers connect to.
Here's the problem: At random times, about 4-10 times a day, we get
flooded
by single network sessions. Every one of the extreme 5i ports will be
solid
orange with collisions. I have one sole machine plugged into an
ordinary
port on the 'core' 5i (no port monitoring, just an ordinary port). I
run
Ethereal on this machine, and it always shows one direction of the
SAMBA
conversation between a test client and server. The MAC address is
UNICAST,
and the IP is UNICAST.
I will kill the test in question, and the summit's will immediately
clear
up.
So WHY do I get this packet on an ordinary switch port, when I am the
only
machine on this port???? This happens in a seemingly random manner,
between
various clients, and various servers, not isolated to one rack or one
summit
5i. We have taken all the 5i's out of the path except 1, and the
problem
occurred. Then we swapped out that summit 5i with another one, and the
problem reoccurred.
The servers under test have multiple ethernet adaptors, but each
adaptor is
on a different class C subnet. The test clients have 1 or 2 ethernet
adaptors, with one ip alias per machine corresponding to each class C
subnet, shared with the server under test.
Has anyone seen this before? Or have some idea as to why this problem
is
occuring? The core switches are all high-end switches, and shouldn't be
filling up their MAC tables. They're all Gigabit as well, so bandwidth
shouldn't be a problem. And if it was, why only a single session being
broadcast?
???
Thanks a MILLION in advance,
-AnthonyM
.
- Prev by Date: Re: switching between dynamic and static IP on wifi laptop?
- Next by Date: Re: How to make the linked led lit in a constant rate on hub?
- Previous by thread: load balancing, fail-over
- Next by thread: Linux cluster queue software
- Index(es):
Relevant Pages
|
