The only difference between SNMPv1 and SNMPv2c polling in InterMapper 3.6 and later is that SNMPv2c polls include variables from ifXTable. If you do not need the 64bit counters (ifHCInOctets, ifHCOutOctets) to monitor extremely fast links, you should get by with SNMPv1 polling. In effect, using SNMPv1 will implement their suggestion of refraining from accessing the ifXTable.
A longer term solution would be for InterMapper to honor their capability MIB statements with respect to ifInMulticastPkts, ifInBroadcastPkts, ifOutMulticastPkts, ifOutBroadcastPkts.
Regards,
Bill Fisher
Dartware, LLC
http://www.dartware.com
On Tuesday, November 12, 2002, at 04:55 AM, Jerome Fleury wrote:
The problem is that when we activate Cisco v2c polling in Intermapper, the Cisco device gets more than 10% CPU load higher. The problem has been identified with the help of Cisco, whose analysis is the following:
CSCdm11645:
ifName is the first variable in the ifXTable, which augments the ifTable. After ifName, the next 4 variables, ifInMulticastPkts, ifInBroadcastPkts, ifOutMulticastPkts, ifOutBroadcastPkts are not supported in the IF-MIB as suggested in the CISCO-IF-CAPABILITIES file. As a result, when a GET-NEXT is done on the ifName variable of the last available interface, the SNMP Research code goes thru' the above mentioned 4 un-supported variables for all the interfaces before it hits the next supported variable, which is ifHCInOctets.1. On a router with thousands of interfaces, this takes up a lot of time, which results in the CPU-HOG messages.
This problem arises because the SNMP Research code has to go thru' the un-supported variables for all the interfaces. Since the SNMP Research code is third-party code, we cannot provide a "fix" in that code. Since there is nothing really wrong with the ifTable code, nothing can be done there either.
As a result, the work-around that could be adopted is to refrain from accessing the ifXTable. If that is done, everything else should be fine.
The problem is quite obvious. Is there any fix possible ?
____________________________________________________________________ Note: To unsubscribe from this mailing list, please send email to: <mailto:InterMapper-Talk-off@;list.dartware.com> Thanks!
