Re: Begin WG Last Call: draft-ietf-rmonmib-raqmon-pdu-09

Hi -

> From: "David B Harrington" <[email protected]>
> To: <[email protected]>
> Sent: Tuesday, January 18, 2005 7:18 AM
> Subject: RE: [RMONMIB] Begin WG Last Call: draft-ietf-rmonmib-raqmon-pdu-09
...
> One of the concerns raised on the MIB Doctor's list was the issue of
> scalability if multi-varbind traps are sent to an SNMP engine in
> realtime from a large number of devices. My personal experience is
> that this may not be scalable. I have had personal experience with a
> deployment that tried to use realtime reporting of events using SNMP,
> and caused a leading SNMP NMS application capable of managing tens of
> thousands of devices to crash (i.e not being able to continue
> operating effectively).
...

In fairness, if I recall your message to the other list correctly, it must
be mentioned that that NMS application did a fork() for each trap
received.  I'm sure the implementors thought that that was a good
idea at the time, but there is nothing in the SNMP protocol or architecture
that demands such an overhead-intensive approach.  I've worked with
systems where the cost of handling an incoming notification was on the
same order as processing a response to a get-next request.

Randy