From: Ryan Morris (ryan@egate.net)
Date: Tue Oct 30 2007 - 10:06:45 ART
Hi everyone,
I'm developing a standardised QoS model for our enterprise. Using Tim
Szigeti's 4/5 class model as a basis, I'm wondering if anyone has real
world experience mapping applications to queues over the wide area.
Here's a breakdown of my current plan:
Voice: 10% priority queue (that's enough to cover all the calls our IP
trunks are licensed for, plus additional room for overhead)
Bulk: 5% Here I'm looking at mapping Lotus Notes and file transfer
traffic. See problem below
Critical: 37% Database apps, call signalling, and identified high
priority web apps (http & https)
Best effort: 25% Everything else
These numbers map fairly well to our measured traffic ratios over time,
except when someone overloads the link with Lotus Notes or file transfer.
Current problem: someone doing a large Lotus Notes replication or a large
file transfer can congest the WAN and affect the performance of the file
transfer traffic and vice versa. If I map Lotus Notes and file transfer
to the same queue, this problem still exists. Note that our Windows file
servers are centralised using WAN accelerators, so there's an interactive
component to the file transfer traffic (i.e. I'm trying to use the Open
function in MS Word) that is sensitive to delay.
Solution: create 2 queues, one for Notes and one for file transfer, to
ensure both are allocated bandwidth.
Question: should we stick with a low bandwidth percentage for these
queues, or should I bump it up to say, 20% each and reduce my critical
data queue? Alternatively, I could map the file transfer traffic to the
critical data queue, due to the interactive component.
Any thoughts or real world experience would be appreciated.
Thanks,
Ryan Morris
CCIE #18953
This archive was generated by hypermail 2.1.4 : Fri Nov 16 2007 - 13:11:19 ART