We have several high end L3 switches with multiple interfaces, but you
bring up a good point thought. I have a discovery filter which should
populate only Cisco devices...Here is a snip of my seed file...
@oid 1.3.6.1.4.1.9.*
@oid 1.3.6.1.4.1.3076.*
I am finding that many other objects are being added to the object
database. If I perform a "ovtopofix -a", then a ton of devices (mostly
hosts) are deleted. Any ideas?
Current
H:\>ovobjprint -S
Number of objects defined in the database: 28866
Total number of fields defined in the database is: 272.
Total number of field values in the database: 230196
Number of Integer fields: 68039.
Number of Boolean fields: 70858.
Number of String fields: 76677.
Number of Enum fields: 14622.
After "ovtopofix -a"
H:\>ovobjprint -S
Number of objects defined in the database: 18546
Total number of fields defined in the database is: 272.
Total number of field values in the database: 209554
Number of Integer fields: 68037.
Number of Boolean fields: 70858.
Number of String fields: 56037.
Number of Enum fields: 14622.
I have SNMP polling as default for a handful of OIDs by design because
there are some public interfaces which I am unable to ping directly.
We have a global network so 50% of the devices have latency over 100 ms
and 25% of the devices are over 180 ms.
Paul
-----Original Message-----
From: Stephen Hochstetler [mailto:shochste@us.ibm.com]
Sent: Monday, June 05, 2006 3:18 PM
To: nv-l@lists.us.ibm.com
Subject: RE: [nv-l] Polling Question
I am wondering about this "managing" 900 items and having 24K items in
the database. Is this really true? Does those 900 network devices
translate to 6000 interfaces? I always thought that I got about 4
objects per interface I was managing.
A demandpoll is handled via SNMP. Do you have SNMP polling as the
default? Do you have to? It is 'lighter' to both the network and also
the NetView server if you are using ICMP for polling. Then for devices
that really need SNMP polling to move those to that type of polling.
This would relieve the SNMP workload that netmon is currently doing and
potentially let it respond to a demandpoll request quicker.
Question -- do you have some nodes that respond slowly? Netmon will only
send out so many 'snmp' requests and wait for responses before
continuing. The symptoms of this is the SNMP queue in netmon gets backed
up but the CPU usage of the NetView server decreases. This is best
looked by by running a netmon trace into a large file, then analyzing it
later. The slow response can be either due to network latency, or a
slowly responding router. At the routers, if they are running high CPU
it will cause a very slow SNMP response because SNMP gets a very low
priority within the router. To these devices doing 'ping' polling can
make a huge difference to NetView.
Stephen Hochstetler shochste@us.ibm.com
International Technical Support Organization at IBM
Office - 512-838-6198 (t/l 678) FAX - 512-838-6931
http://www.redbooks.ibm.com
|