4000M Procurve - SNMP queries / Not enough Buffers

[Booker Apelin]

I have 3 hp 4000M switches that I need to have monitored via snmp by multiple sources. Multiple sources I mean, mrtg, cricket, intermapper, and in house perl scripts. In my research I am finding that I am exhausting buffers whenever I do snmp queries. These are the same buffers for packets so the major side affect is that packets start dropping on my netowrk because snmp responses are exhausting my buffers. I need to get a recommendation on modifying my switch's configuration or hardware upgrades so that it can handle numerous snmp queries.

Below are some info on what I have:

SWITCH A:
6 - 10/100 cards
(28 ports in use)
2 - Gig/sx card
Firmware: C.09.16
SWITCH B:
7 - 10/100 cards
(30 ports in use)
1 - Gig/sx card
Firmware: C.09.16
7 - 10/100
(35 ports i use)
1 - Gig/sx card
Firmware: C.09.16

The 3 switches are bridged together via the fiber (gig/sx card):
B <-> A <-> C

We had almost 25 vlans configured for each switch but have reduced that to 20. That has helped so we think since there are less vlan's that has freed up some memory and also increased the number of buffers.

Here's and example of my testing. From the Status and Counters screen of the switches I have:

Switch A,B and C have the same memory displayed:
Memory -
Total : 7,367,520
Free : 4,424,112

For Buffers Switch A has:
Packet - Total : 382
Buffers Free : 139
Lowest : 0
Missed : 14

Switch B:
Packet - Total : 382
Buffers Free : 139
Lowest : 39
Missed : 0

Switch C:
Packet - Total : 382
Buffers Free : 138
Lowest : 4
Missed : 0

I recently power cycled the switches so the counters have reset but previous to the power cycle, all switches had a high number of missed in it's buffers and lowest was a value of "0". As a test, I have a perl script on one of my unix servers that uses snmpbulkwalk. This script just get's the status of the ports on the switch and what's on the port.

When I run this script, on Switch A, Free Buffers will drop to almost 1 or to 0. When it drops to 0, I start to get missed packets, which I believe in turn affects all ethernet traffic and starts dropping packets. On the other switches the you can see that the buffers get low, Lowest : 39 and Lowest : 4 so on switch B and C I haven't had a packet loss problem, but that's because many of my monitoring tools are off. I don't have mrtg server or my cricket server quering the switches and nor do I have all my probes from intermapper monitoring the switch, just a small subset.

I'll be glad to provide any other information on my setup. Basically I want to be able to do more snmp quering but the hp's seem to be limited on buffers.

Booker
hp-forum@lava.net