Some limits of netmon in large networks
ITEM: RS4000016657
**************> QUESTION level 1 --> level 2 SPECIALIST <**************
My customer is monitoring large network with NetView/6000 V2, and
wondering if netmon can't poll every nodes in the polling interval
which they customized at SNMP configuration panel.
Are there any limits to netmon's status polling?
For example, how many nodes can be polled in one second?
what will occur to the polling timer if netmon can't poll some nodes in
the polling interval?
Sorry for my complex question but please let me know about the detail
mechanism of netmon's status polling...
Thanks a million in advance.
**************> ANSWER level 2 --> level 1 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN USA
====> ASGN: NHL AT WTSCPOK ======================== DATE:951019
TIME:0851
Received by Raleigh ITSC and assigned to HARIKIAN.
**************> ANSWER level 2 --> level 1 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN USA
====> ASGN: HARIKIAN AT WTSCPOK ======================== DATE:951025
TIME:1220
The question has been re-assigned to MCGREGOR.
**************> ANSWER level 2 --> level 1 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN USA
====> RESP: MCGREGOR AT WTSCPOK ======================== DATE:951026
TIME:0950
The way netmon polling works is:
1. It attempts to poll every IP interface using a simple ICMP echo (a
ping) at the interval that you define in the SNMP configuration
panel
2. If an interface does not respond, it uses the retry scheme (also
defined in SNMP Configuration). For each retry the timeout value is
doubled, for example:
Poll<-- t -->Poll<---- 2t ---->Poll<-------- 4t -------->Set Node down
This would be the case for a timeout value of "t" seconds and a
retry count of 2. For the standard configuration of t=0.8 seconds
and
a retry count of 3, the total timeout period is .8 + 1.6 + 3.2 + 6.4
seconds - ie 12 seconds.
3. netmon will only allow a certain number of polls to be outstanding
at one time. In Version 2 the limit was originally 3 polls. It been
increased to 10 in V3 and V4, but I am not sure whether V2 PTFs have
increased this. In other words, if netmon sends out three pings, and
gets no response before the timeout interval expires, it won't try
to send out pings to other interfaces. This ultimately puts a limit
on the rate at which netmon can send out pings. As your network
reaches a certain size, it is possible that netmon may not be able
to ping all interfaces in the network in your polling interval
(there
is no way to say PRECISELY if this will happen - it depends on how
fast the nodes in the network respond).
If you want to increase the number of outstanding
polls, do the following:
a. Make a backup copy of /usr/OV/lrf/netmon.lrf
a. Open the file /usr/OV/lrf/netmon.lrf in an editor
b. On the second line, after the second ":", you should see a
"-P"
c. After the "-P", and before the third ":", type in the
following text
, -q n, -Q n (where "n" is the number of outstanding polls
you want - I woudl suggest changing to 10 or 20 at first)
d. After you finish, the line should look like this:
OVs_YES_START:ovtopmd,trapd,ovwdb:-P, -q 20, -Q 20:OVs_WELL_BEHAV
ED:15:
e. Type the following command:
/usr/OV/bin/ovaddobj netmon.lrf
f. Stop and restart netmon
ovstop netmon
ovstart netmon
The only downside to this is that editing this file by hand
eliminates your ability to change the netmon configuration with
SMIT. Any further changes to netmon (like adding a seed file, for
instance) would have to be done by hand rather than with SMIT.
From the above description, you can see that if netmon gets behind with
its polling it will just cause subsequent polls to be delayed, which
means that some nodes may get an extended polling interval. The best
way to address this problem is to look very hard at what nodes you are
polling. In most networks there is a very small number of really
imoprtant nodes. For those you want a short polling interval so that
you know as soon as something goes wrong. For the majority of nodes
you really do not need to know if they are up or not. for those you can
either set a long polling interval or define them as unmanaged so that
they are not polled at all.
Finally, you should try to get the customer onto V3 or V4. The internal
operation of netmon has been improved considerably over the past two
years, and not all of the improvements have been applied
retrospectively.
**************> QUESTION level 1 --> level 2 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN JAPAN
Thank you for your answer˘
Can I ask one more quetion on netmon trace?
I found the following entries in my netmon trace.
ICMP message from: x.x.x.x
unexpoectec ICMP message 3 from x.x.x.x
I understand these icmp echo is unrelated to netmon and there is no
harm
to netmon's performance. Is it correct?
Moreover I couldn't find any document about netmon trace format. If
there
is, please let me know.
Thanks in advance.
**************> ANSWER level 2 --> level 1 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN USA
====> ASGN: NHL AT WTSCPOK ======================== DATE:951030
TIME:0921
Received by Raleigh ITSC and assigned to SHOGREN.
**************> ANSWER level 2 --> level 1 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN USA
====> RESP: MCGREGOR AT WTSCPOK ======================== DATE:951030
TIME:1729
You are quite right - the "unexpected ICMP message" is just netmon
intercepting a ping response that it did not originate. There is no
impact from this.
The format of the trace is not documented anywhere, but the individual
messages are now in the NetView for AIX Messages manual. This is new in
Version 4.
**************> QUESTION level 1 --> level 2 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN JAPAN
Thank you for your note˘
I have now two more questions about netmon polling.
Sorry but could you please tell me once again?
1. I heard that netmon has a polling list for its status polling.
Is the order in the list usually notchangeable?
2. If the answer for the above is yes, I wonder if the 'same' nodes
aren't be polled when netmon can't poll some nodes in a defined
polling interval. Is there a possibility that such a phenominan
occures?
Thanks a million in advance.
**************> ANSWER level 2 --> level 1 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN USA
====> ASGN: NHL AT WTSCPOK ======================== DATE:951031
TIME:1541
Received by Raleigh ITSC and assigned to MCGREGOR.
**************> ANSWER level 2 --> level 1 SPECIALIST <**************
====> THIS TEXT HAS BEEN ENTERED BY IBM IN USA
====> RESP: MCGREGOR AT WTSCPOK ======================== DATE:951031
TIME:1621
The polling list is internal to netmon, and cannot be changed. If
netmon
gets behind with its polling, it SHOULD just go through the list in
order. This means that polls may be delayed, but there should not be
any cases in which nodes are not polled at all, or other nodes are
polled out of order. However, I do recall some problems of this type
with netmon in V2. If you find evidence of polls happening out of order
you should report it as a defect.
WWQA: ITEM: RS4000016657 ITEM: RS4000016657
Dated: 10/1995 Category: AIX6000
This HTML file was generated 99/06/24~12:43:06
Comments or suggestions?
Contact us