Hello,
I have a couple of questions regarding performance measurements of
interfaces on Cisco 7600 router (and other models) and that is the
reason for lengthy message:).
Following OID-s can be used for measurements of inbound traffic (there
are equivalent OID-s for outbound traffic):
1. 32bit counter for number of octets: IF-MIB::ifInOctets.L2
2. 64bit counter for number of octets: IF-MIB::ifHCInOctets.L2
3. 32bit integer for 5min EMA:
OLD-CISCO-INTERFACES-MIB::locIfInBitsSec.L2
4. 32bit counter for number of octets: IF-MIB::ifInOctets.L3
5. 64bit counter for number of octets: IF-MIB::ifHCInOctets.L3
6. 32bit integer for 5min EMA:
OLD-CISCO-INTERFACES-MIB::locIfInBitsSec.L3
7. 32bit counter for number of IP octets:
OLD-CISCO-INTERFACES-MIB::locIfipInOctets.L3
L2 and L3 represent ifindex numbers of Layer 2 and Layer 3 interfaces.
Layer 3 interface is VLAN interface (for example Vlan10), and Layer 2
is GigabitEthernet interface:
interface GigabitEthernet1/10
description test
load-interval 30
switchport
switchport access vlan 10
Both interfaces have load-interval set to 30 seconds, but I think this
doesn't have any influence on SNMP polling.
According to OLD-CISCO-INTERFACES-MIB locIfipInOctets represents "five
minute exponentially-decayed moving average of input bits per second".
I have done tests of polling with very short interval (100 ms) just to
check how often are these counters updated. From my results, they are
updated every 10 seconds, so polling faster than that doesn't makes
sense. Although ifInOctets and locIfipInOctets for Layer 3 interface
have small drift, they change few hundred octets with every reading,
but every 10 seconds they have normal update as other counters.
And now come questions:).
Counters #1 and #2 are very consistent (not even one octet of
difference), while counter #3 gives slightly different readings. Since
it is MA this makes sense. I'm not sure if it is still 5 minute
average, regardless of load-interval setting. Is this counter always
calculated on 5 minute interval or it depends on load-interval?
Counters #4 and #5 are not consistent with #1 and #2, I guess they
don't count L2 overhead, and there is also small difference between
themselves. Difference between Layer 2 and Layer 3 counters is around
5%, which is consistent with Ethernet overhead for packets with size
around 400-500 bytes.
They are also updated on different time points, #1 and #2 are always
updated in same 100 ms point, but #4 and #5 lag or lead few hundred ms
after or before them and they are not updated simultaneously.
What I don't know is why are they not updated on same time points as
Layer 2 counters. Does every interface has it's own internal timer?
Why is there a difference between readings for 32bit and 64bit
counters for same Layer 3 interface? Do they also have different
timers?
Counter #7 is updated at same times as counter #4 but with very slight
differences. Counter #7 gives on average around 7 octets less when
compared to counter #4. What is cause of these differences and why do
these two counters have slight drift (see above)?
And finally #6 is updated at his own time points (but every 10
seconds) and again gives different readings from counter #3. This
equals with Layer 2 overhead.
Thanks for any answers and if anybody has other information regarding
these measurements, please share them.
