nv-l
[Top] [All Lists]

RE: [nv-l] SNMP performance

To: nv-l@lists.us.ibm.com
Subject: RE: [nv-l] SNMP performance
From: "Evans, Bill" <Bill.Evans@hq.doe.gov>
Date: Fri, 12 May 2006 18:48:46 -0400
Delivery-date: Fri, 12 May 2006 23:49:30 +0100
Envelope-to: nv-l-archive@lists.skills-1st.co.uk
Reply-to: nv-l@lists.us.ibm.com
Sender: owner-nv-l@lists.us.ibm.com
Thread-index: AcZ1+kDIlRCxaMW+QU6M7KRwi8f2cQAFveJQ
Thread-topic: [nv-l] SNMP performance
I don't think that's a very practical design.  The default times for
network polling are five minutes in Unix and Linux systems and twenty
minutes in Windows.  I've implemented special exceptions on a one minute
frequency for my organizations gateways to the internet.  Milliseconds
are another order of reality.  

That said, a device which sends traps on a ten millisecond basis would
be considered a sick box in most installations. That's six thousand
transactions per minute!  NetView MIGHT be able to handle one box at
that rate but I would NOT plan on that happening.  There's too much
processing involved with each trap.  A SHORT burst at that rate should
be no problem on a fast Unix/Linux machine since the burst would be
queued. I'm talking about a burst of a few seconds.  

The easiest way to get an answer is to try it.  You won't break
anything.  Put up NetView on the machine you would use and start pumping
traps from another task.  Here is a sample NetView custom trap.  Put it
in a shell script and issue it for five minutes in a loop with "date %T"
and see how fast you can get it to send to NetView on the same machine.
 
        /usr/OV/bin/snmptrap ${hostname} .1.3.6.1.4.1.2.6.3 ${hostname}
\
          6 1999 0 .1.3.6.1.4.1.2.6.3.1.1.1.0 OctetString "Normal"

When the time is up, browse /usr/OV/log/trapd.log to see the record.  If
you want finer tracing and to really load the system enter "netmon -M
32" before running the test.  If your NetView daemons crash don't be
surprised -- you just found the system limits.  The most likely result
is the message which indicates you exhausted the queue of incoming traps
and dependent daemons disconnected.  

Let us know your results.

Bill Evans

-----Original Message-----
From: owner-nv-l@lists.us.ibm.com [mailto:owner-nv-l@lists.us.ibm.com]
On Behalf Of Xianan s Zhang
Sent: Friday, May 12, 2006 3:28 PM
To: nv-l@lists.tivoli.com
Subject: [nv-l] SNMP performance

Hi,

May I ask a question about the snmp performance? How frequently do SNMP 
manager and agents communicate usually in practical systems?

We are thinking about building a tool in which the agent can report data

back to the manager every 10 milliseconds.  Can SNMP implementations 
support communication such frequently? 

Thanks!

Shannon



<Prev in Thread] Current Thread [Next in Thread>

Archive operated by Skills 1st Ltd

See also: The NetView Web