From my SNMP guy:

UDP port 161 is the default port on which SNMP listens for incoming
snmp get/set requests.  My guess is that somewhere on the network is
an SNMP manager that is sending requests to that host.  That being
said, it looks to me like the host's SNMP master agent is not reading
from the port and that is why the receive q is filling up.  This could
happen if a request results in snmpd getting into a deadlock situation.

I think I'd start trying to figure out the problem by using tcpdump to
see if I could determine if, in fact, there are SNMP requests being sent
to this host and if so, where they are coming from.

Regardless of whether or not there are SNMP requests, enable snmpd
logging of everything and check that information when the receive q
starts to fill up.

Altermately, tcpdump may tell you that the host is being flooded with
SNMP requests.  If this is the case, it may explain the system performance
impact because for the most part, SNMP has to query for information
from the kernel and/or kernel modules via ioctl/etc.  If the host is being
flooded with SNMP requests, it's possible that it may be doing nothing
but servicing those requests.


At 02:44 PM 3/1/2005, Biggs, Eric J [ITS] wrote:
We are using RHEL AS 3.0 (update 3) on a z990, under z/VM 4.4 and have
been experiencing a strange problem related to snmp and various other
programs and utilities. Take a look at this netstat output for port 161:

Active Internet connections (only servers)
Proto Recv-Q Send-Q Local Address           Foreign Address
State
udp   130928      0 0.0.0.0:161             0.0.0.0:*

Notice the huge amount of data in Recv-Q.  This situation just builds
over time.  We haven't been able to determine anything that might be
causing it.  Here's the strange thing:  Other programs and utilities get
"stuck" when they are used whenever this "queue" exists for port 161.
If we restart the snmpd service, this queue clears up and the other
programs free up.  As an example, if I enter a sudo command it just
hangs the curser.  I have to close the session.  Same thing with the
ifconfig command, it just hangs.  This was also observed using Red Hat's
"up2date" command for accessing Red Hat Network.  The two issues are
obviously related because I can open another session (after I've hung
one of those commands), restart the snmpd service, then go back to the
first window and the command completes.  Weird.  There are probably
other commands that are impacted too, we just haven't found them yet.

The vendor support that we pay for (I won't mention any names) has not
been able to shed much light on this issue.  Does anyone have any ideas,
or have you experienced a similar situation?

Eric Biggs



----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390


----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to