We use SNMP to graph several key metrics about our Nexus 5000/2000 environment. There is a really rich MIB of values to choose from.
Consider the standard IF-MIB below, normally we just graph the high-capacity counters ifHCInOctets and ifHCOutOctets unless we have problems.
IF-MIB::ifName.527370112 = STRING: Ethernet112/1/31
IF-MIB::ifInMulticastPkts.527370112 = Counter32: 581867
IF-MIB::ifInBroadcastPkts.527370112 = Counter32: 28822
IF-MIB::ifOutMulticastPkts.527370112 = Counter32: 482529429
IF-MIB::ifOutBroadcastPkts.527370112 = Counter32: 17470234
IF-MIB::ifHCInOctets.527370112 = Counter64: 38076297789952
IF-MIB::ifHCInUcastPkts.527370112 = Counter64: 12227880172
IF-MIB::ifHCInMulticastPkts.527370112 = Counter64: 581868
IF-MIB::ifHCInBroadcastPkts.527370112 = Counter64: 28822
IF-MIB::ifHCOutOctets.527370112 = Counter64: 78374736556136
IF-MIB::ifHCOutUcastPkts.527370112 = Counter64: 14210055971
IF-MIB::ifHCOutMulticastPkts.527370112 = Counter64: 482529477
IF-MIB::ifHCOutBroadcastPkts.527370112 = Counter64: 17470239
IF-MIB::ifLinkUpDownTrapEnable.527370112 = INTEGER: enabled(1)
IF-MIB::ifHighSpeed.527370112 = Gauge32: 10000
IF-MIB::ifPromiscuousMode.527370112 = INTEGER: false(2)
IF-MIB::ifConnectorPresent.527370112 = INTEGER: true(1)
IF-MIB::ifAlias.527370112 = STRING: NFS Server ixgbe0
IF-MIB::ifCounterDiscontinuityTime.527370112 = Timeticks: (0) 0:00:00.00
When we have problems, we graph more diagnostic values from the Etherlike-MIB. But we recently discovered that some of these values never increment. In this case specifically, we're talking about dot3HCInPauseFrames and dot3HCOutPauseFrames.
EtherLike-MIB::dot3StatsIndex.527370112 = INTEGER: 527370112
EtherLike-MIB::dot3StatsAlignmentErrors.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsFCSErrors.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsSingleCollisionFrames.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsMultipleCollisionFrames.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsSQETestErrors.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsDeferredTransmissions.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsLateCollisions.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsExcessiveCollisions.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsInternalMacTransmitErrors.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsCarrierSenseErrors.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsFrameTooLongs.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsInternalMacReceiveErrors.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsSymbolErrors.527370112 = Counter32: 0
EtherLike-MIB::dot3StatsDuplexStatus.527370112 = INTEGER: fullDuplex(3)
EtherLike-MIB::dot3InPauseFrames.527370112 = Counter32: 0
EtherLike-MIB::dot3OutPauseFrames.527370112 = Counter32: 0
EtherLike-MIB::dot3HCInPauseFrames.527370112 = Counter64: 0
EtherLike-MIB::dot3HCOutPauseFrames.527370112 = Counter64: 0
Consider the output from this command:
# show interface flowcontrol | exclude "off off 0 0"
--------------------------------------------------------------------------------
Port Send FlowControl Receive FlowControl RxPause TxPause
admin oper admin oper
--------------------------------------------------------------------------------
Po213 on on off off 0 3462769
Po233 on on off off 0 1224473
Po234 on on off off 0 3065991
Eth111/1/1 on on off off 0 108658400
Eth111/1/2 on on off off 0 131389
Eth111/1/25 on on off off 0 179289
Eth111/1/31 on on off off 0 1043047
Eth112/1/1 on on off off 0 38446983
Eth112/1/2 on on off off 0 3331539
Eth112/1/25 on on off off 0 2900259
Eth112/1/31 on on off off 0 181426
While the output of the CLI shows several interfaces that have high numbers of Tx Pause frames recorded, and any non-zero value should have been reflected in the SNMP MIB (regardless of whether or not the "clear counters" command has been issued) all of the appropriate MIB values are zero. This has made correlating the Tx Pause frame with either the packet rate or the traffic volume almost impossible.
We really need these values when things go wrong. But isn't that always the case. When things go right, who needs all this data? But when you're trying to pinpoint the cause of a performance bottleneck, every statistic you can gather casts light on the problem.
