[nsp] characterizing router performance

From: Priscilla Oppenheimer (cilla@priscilla.com)
Date: Thu Feb 26 1998 - 17:49:32 EST


Hello nsp,

I'm writing a book on Top-Down Network Design. I'm working on a chapter
that briefly covers analyzing the existing network performance before
selecting solutions for the new design.

I'd like to include in the book some real world advice on analyzing
existing routers, so that people such as yourselves can learn some basic
techniques for characterizing the health of a router. (No gory details,
since the focus of the book is design not performance analysis.)

I'm looking for insight into your method and process for checking Cisco
routers. Do you use show commands, SNMP, other tools? How often do you
check values? And, most importantly, at what point do you consider
performance values problematic?

Can you help by answering some questions?

In the "real world" do you track the following:

CPU utilization of a router?

   MIB variable avgBusy1 (1 minute average of the CPU busy percentage)
   MIB variable avgBusy5 (5 minute average of the CPU busy percentage)
   How often do you check (i.e. what is your process?)
   At what point does avgBusy1 concern you?
   At what point does avgBusy5 concern you?

Interface statistics?

   Queue drops?
   MIB variable locIfInputQueueDrops?
   MIB variable locIfOutputQueueDrops?
   How often do you check (i.e. what is your process?)
   At what point do you consider input Q drops a problem?
   At what point do you consider output Q drops a problem?
   Ignoreds?
   MIB variable locIfInIgnored?
   At what point do you consider Ignoreds a problem?

Buffer statistics?

   Misses for elements, small, medium, etc. buffer pools?
   MIB variables bufferElMiss, bufferSmMiss, etc.?
   How often do you check (i.e. what is your process)?
   At what point do you consider misses a problem?
   Buffer allocation failures?
   MIB variable bufferFail?
   At what point do you consider failures a problem?

I know this isn't a complete list, but that's because I have a good feel
already for characterizing bandwidth utilization and frame errors. The
questions above are ones I could use the most help with (though any generic
advice is also welcome.)

Thank-you very much. If you can help, I can put your name in the
acknowledgements section of the book and send you a free copy when it is
published!

Priscilla Oppenheimer

__________________________________
Priscilla Oppenheimer
Phone 541-482-5685
Fax 541-488-1708
Web http://www.priscilla.com



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:15 EDT