From: Scott Vermillion (scott_ccie_list@it-ag.com)
Date: Mon Mar 03 2008 - 17:48:49 ARST
You've got it Sadiq!
If a packet can be placed directly on the TxRing HW FIFO queue, it is. If
not, it has to go in the software output queue where your congestion
management stuff is applied (actually, I think conceptually, it's applied
*between* the SW and the HW queue, but that's splitting hairs). This is the
basic definition of congestion. This is why when you configure QoS on an
interface, the TxRing is actually truncated from its default depth, to make
it more likely congestion will cause the policy to be invoked. Or so I
understand things from "Inside Cisco IOS Software Architectures," which is
admittedly a dated reference, but still more or less valid I believe...
Regards,
Scott
-----Original Message-----
From: Sadiq Yakasai [mailto:sadiqtanko@gmail.com]
Sent: Monday, March 03, 2008 12:41 PM
To: Scott Vermillion
Cc: Spolidoro, Guilherme; Gaurav Prakash; groupstudy groupstudy
Subject: Re: LLQ
Scott,
You are absolutely right there. This is always one of those areas in
which one needs to hypothesize (if this word exists :)) i guess.
It basically comes down to the definition of congestion on the interface.
In the second case (B), they wld definately be policing themselves
even without congestion.
So, when is the interface "congested" again? When they packets start
filling up the interface queue? If so, to what limit?
Mayb i need to check with my Odom and refresh me mind here.
This archive was generated by hypermail 2.1.4 : Tue Apr 01 2008 - 07:53:52 ART