Per VLAN Stats on MSFC2 - Complaints from the Field

Hey all,

This one is a weird one. I apologize if this is a bit off topic.

As everyone is probably aware, the Cisco 6500/7600 line is unable to
provide per-vlan I/O statistics on routed interfaces (ie, a "show int vlan
xxx" has meaningless numbers in the I/O and error fields at the end).
MIB tables also fail to provide meaningful data. You CAN get some L2 VLAN
data, but that doesn't help you when you need to know what's going through
the router interface.

We've be going around and around with the vendor for a while now about how
this makes it pretty useless for traffic analysis, and even showed them
the RFC that requires that the information be made available to call it a
router. Their latest tactic is to claim that "nobody else in the industry
is concerned about this shortcoming".

For us, we've been collecting routed interface stats going back all the
way to 1996 and with our new gear we've been sold we find that our graphs
are pretty empty and we get no reasonable information about how much
traffic is passing through the L3 interfaces.

So, here's the question I have for anybody out there dealing with this
hardware. Is this shortcoming an issue for how you run your network?
Have you asked Cisco to fix it? My feeling is that a lot of people find
it to be an issue and simply accept it is yet another broken cisco thing.

Please let me know privately. I want to talk into our next meeting with
some printed testimonials from "real people" that this box is plain
broken. Let me know as well if you don't want me to use your name and/or
organization in my report.

Thanks for your time.

- Robert Hayden
University of Wisconsin Madison

This is because in 1996 you were likely not dealing with 'Switch Routers'; today's 'routers' perform some form of flow switching/caching, meaning once the traffic enters the VLAN routed interface and an appropriate path is found it is sent down the the Layer 2 fabric. This can be circumvented by disabling MLS on your fabric, but will result in all of your packets being process switched, inherently increasing the CPU load of your MSFC/CPU.

Depending on your configuration, your SVI information can be coupled with the Layer 2 SNMP statistics (at least for I/O) to provide more accurate numbers.

I apologize if I have missed something, but I'm assuming this is what you're alluding to.


Anthony Cennami

Robert A. Hayden wrote: