|
Joe
and Leslie,
Thanks
for your advice. I'll do some further investigation and testing, especially to
avoid collecting those "non-reply" and "non-response" objects.
One
other "problem" appearing in the trace file:
"MIB
cpmCPUTotal5min, (and ciscoMemoryPoolFree) on 'hostname' gave
type Gauge, expect TIMESTICKS (due to mib.coerce
file?)"
1) I
did not specify anything in the coerce file and checked at Cisco site that CPU
and Memory should be type Gauge (32). Why they expect
TIMESTICKS?
2)
Does this error have impact on the quality of data collection (e.g. delay
etc.)?
Regards,
David
Startup of
snmpCollect may be slow. Turn on the tracing and watch what it is doing at
startup, so you know whether to worry or not.
I've found snmpCollect to be pretty efficient at
collecting, and at minimizing its impact on the devices by grouping stuff
together. Where you may have trouble is with those per-interface values on
devices with lots of interfaces. There is a parameter for the snmpCollect
daemon for the maxpdu that controls how much data it will request at once. It
defaults to 100 somethings. I've sometimes changed it to 50 somethings, so
snmpCollect breaks the request into smaller packages. This avoids loss of data
caused by the device refusing to deliver too-large responses.
There is also, under the
Options....SNMP Config, the timeout and retries settings. I know this
applies to snmp requests from other parts of Netview, but I have never
been sure whether it applied to snmpCollect or not.
You will also have trouble with some of those interface
counters if the interfaces are very high speed. Netview currently will only
collect 32bit values (Counter32), and for gig interfaces, the values wrap much
too quickly. So take a look at which instances you really need, and what the
rate of flow really is. There is no point collecting it if it is bad data.
Look for sub-interface instances with lower rates of flow and see if that will
give you what you need.
When you
update snmpCollect configuration via the gui, it updates
/usr/OV/conf/snmpCol.conf and stop/starts snmpCollect. You can edit that file
manually. If you find that you want to collect a variety of different
interface instances on each device, you could generate that file
programmatically. I'm suggesting that with large numbers of devices, entering
*.*.*.* or Routers just because it is easy is not always the best choice. Try
making fancy entries via the gui and then check the results in snmpCol.conf.
Then write a little script to generate the repetitive parts.
Cordially,
Leslie A. Clark
IBM Global
Services - Systems Mgmt & Networking
"Liu, David"
<david.liu@eds.com>
Sent by: owner-nv-l@lists.us.ibm.com
01/11/2005 03:53 AM
|
|
To
| "'nv-l@lists.tivoli.com'"
<nv-l@lists.tivoli.com>
|
|
cc
|
|
|
Subject
| [nv-l] NV tuning for
Data collection |
|
Hi list,
I've been reading nv-l archives for quite some
times and benefit from them.
Now I post my first question to get your
advice/help.
We are collecting quite a lot data every 15 mins (supposed
to), but till now
only part of the collection happened (everyday less than
half i.e. 40
collections per definition per device, sometimes even null).
My basic
question is: can NV handle that many collections? Because when I
suspended
more than half of the collections (interface data). It seemed
working fine.
If yes, how can I tune the NV? Where can I find the document
for the tuning
(snmpCollect daemon settings)?
Here's some basic
info.
NV 7.1.2 on Solaris 2.8.
Data collection of about 600
devices with
1) SysUptime for all of them
2) cpmCPUTotal5min for
400 devices (cisco)
3) ciscoMemoryPoolUsed for 400 devices
4)
ciscoMemoryPoolFree for 400 devices
5) ifInUcastPkts, ifOutUcastPkts,
ifInErrors, ifOutErrors, ifAdminStatus,
ifOperStatus, ifLastChange,
ifInOctets, ifOutOctets, ifInDiscards,
ifOutDiscards, ifInNUcastPkts,
ifOutNUcastPkts for about 200 devices with
lot of interfaces
6) Some
latency data for 10 routers
Our current snmpCollect (daemon)
settings:
Defer time:
60
Max PDU:
50
Config check interval:
1440
Max concurrent SNMP sessions: 50
Verbose trace mode :
Yes
Polling interval for nvcold:
60
Thank you in
advance.
Regards,
David
|