Question 6: You will need to be careful about a few things.
I recently did a study of the "burden" that rapid polling does on CPU
performance. SNMP daemons responding to queries place a higher burden (via
the number of lines of code) than, for example, simple system calls (were
that information available, such as octet counts in/out an interface). It
is quite easy to place a 1% load on a host, especially if it is an older,
slower processor. Higher loads are also easy. Asynchronous polling works
great, but it has the potential to get you into trouble for this reason.
For a point of reference, a SPARC 10 would bog down (>1% CPU burden) with
more than one or two queries per second. At the other end of the spectrum,
I would see burdens appear on an Ultra-60 only after 500 to 1000 queries per
second were sent to the host.
I do not have figures on the performance regarding polling a whole tree
(your whole row with a single PDU) - my testing only involved individual
leaves/OIDs.
In my situation, I had 30 to 50 hosts to poll in under one second timing. I
polled each for one OID, processed, then did it again with another OID, etc.
I maintained close watch over how often I was asking questions of each
host/architecture type.
Keith Plourd
P. O. Box 151
16 Tuckaway Shores Road
Nottingham, NH 03290
(603) 895-1438 (603) 498-2639 (Cell)
-----Original Message-----
From: owner-tkined@ibr.cs.tu-bs.de
[mailto:owner-tkined@ibr.cs.tu-bs.de]On Behalf Of Szokoli Gabor
Sent: Tuesday, February 15, 2000 5:21 AM
To: tkined@ibr.cs.tu-bs.de
Subject: [tkined] High volume SNMP poller
Hello!
I'm working on a generic SNMP poller, for network performance statistics
and a lot more.
The number of nodes forces me to optimise for network utilisation.
I'm using scotty2.1.10
That's where my questions start:
1,Is that the latest version? How "alpha" is the 3.0.0?
Are there major changes?
Should I be using that, or the stable one?
2,I am looking for extensions, examples, mostly in the field of SNMP
polling. Is there a more detailed index for such contributions than the
ones I've found so far?
3,I want Tnm's get (n) and getnext (n) NOT to format types like TimeTicks
in sysUptime, I know how to format them back, but why waste those
CPU cycles?
I've got hundreds of hosts to poll for thousands of variables.
4,I seem to have Tcl problems with scotty: the program does not exit when
the script ends, as if I used vwait or wish.
5, This one is pure Tcl: How can my script be avare of signals sent to it?
Perl assigns procedures to signals, can Tcl do that?
6,I wrote a "smart SNMP walk", one that retrieves multiple subtrees
(tables), with the minimum number of UDP packets possible
(assuming the number of rows to be unknown, like in ifTable):
each snmpnext carries multiple variables, a whole row.
(I need to prepare it for the tooBig error yet, but I won't use it to
get more than 63 columns anyway )
It is about 2 pages with loads of commnets, and I'm new to TcL.
I mean it's easy, even I did it, and it interfaces more or less
like snmpwalk: callback script, % things regsubst-ed and all.
It works asyncronously for now, you may block with vwait.
How come nobody does it this way?
What fatal problem did I overlook?
HP's snmpwalk, the UCD example snmpwalk, even Tnm's snmpwalk
wastes one PDU per variable in each "row" of the tables.
What's wrong with sending a whole row in a single PDU?
(Should I post it here?)
whoa, that's a big post to start with, I just subscribed and promis not to
talk this much later on :-)
Thanks for any answers or comments:
Szokoli "Szocske" Gabor
-- !! This message is brought to you via the `tkined & scotty' mailing list. !! Please do not reply to this message to unsubscribe. To subscribe or !! unsubscribe, send a mail message to <tkined-request@ibr.cs.tu-bs.de>. !! See http://wwwsnmp.cs.utwente.nl/~schoenw/scotty/ for more information.-- !! This message is brought to you via the `tkined & scotty' mailing list. !! Please do not reply to this message to unsubscribe. To subscribe or !! unsubscribe, send a mail message to <tkined-request@ibr.cs.tu-bs.de>. !! See http://wwwsnmp.cs.utwente.nl/~schoenw/scotty/ for more information.
This archive was generated by hypermail 2b29 : Mon Jan 08 2001 - 15:27:35 MET