From: Chris Lewis \(chrlewis\) (chrlewis@cisco.com)
Date: Sat Aug 27 2005 - 11:25:41 GMT-3
I do not believe it is possible to answer the question as phrased.
There have been two areas of study related to this.
The first looks at what level of utilization yields low latency and
jitter.
Anna Charny did a worst case study which suggested 15%
At NANOG, Casner showed an empirical study of a backbone provider
showing that at 50% utilization low latency and jitter was assured
Bonald did a study that suggested this could be pushed to 75%
Thomas Telkamp adds to this by showing the utilization attainable with
good results is related to link size, as bigger links offer more
aggregation and therefore the total traffic profile is less bursty than
say a T-1, with results that on 10Gig links, 90% utilization is possible
while good results are obtained.
Now if this is just TCP traffic, the seminal article is by Mathis, the
most relevant quote here is "Bandwidth does not define the attainable
TCP throughput, that is dependent upon many factors including:
round-trip delay, loss rate, TCP implementation, RED/Tail-Drop
implementation"
At T-1 levels the answer you seek is highly dependant upon the
following:
Specific loss and latency targets
The number of flows over the interface
The distribution of those flows, in terms of whether they all burst at
the same time
The time interval you choose to measure the utilization over
The traffic profile, any queueing or congestion avoidance mechanisms
implemented
If I were you, I'd look at what performance my applications need in
order to work properly as defined by loss, latency and jitter, then use
IP SLA (previously Cisco Service Assure Agent) to monitor that
performance on an end to end basis (looking at utilization of one link
in the end to end path can be pretty pointless, as these figures only
matter on an end to end basis) to tell you when those figures are are
not being met, rather than trying to work out what utlization level you
can go up to on any given link, as that is almost impossible to figure
out with any accuracy.
Chris
References cited here and if you want to go further,
http://citeseer.ist.psu.edu/ is great
[BONALD]
"Statistical Guarantees for Streaming Flows Using Expedited Forwarding",
Thomas Bonald, et al, INFOCOM 2001
[CASNER]
"A fine-grained view of high-performance networking", Stephen Casner et
al, Packet Design, NANOG 22
[CHARNY]
"Delay bounds in networks with aggregate scheduling", Anna Charny,
Jean-Yves Le Boudec, April 14 2001
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
Scott Smith
Sent: Friday, August 26, 2005 5:27 PM
To: groupstudy
Subject: T1 Utilization Question
I'm not sure how to word this question but I'm going to try anyway...
Assumptions: Full T1, 80% of the traffic is a combination of telnet and
www. The other 20% is a combination of Exchange, Netware, other.
At what point, generally speaking, in utilization percentage do the
users start to feel the effects? 50%, 60%, higher? Or maybe lower?
At what point in utilization does a T1s efficiency drop to the point the
users notice? If at 0% utilization the circuit is 100% efficient then is
the circuit 0% efficient at 100% utilization? Is it linear or is the
circuit 50% efficient at 80% utilization?
Does anyone understand what I'm trying to find out? I know there are
several true experts in this group so I hope one of you can figure out
what the hell I'm asking for. :-)
Thanks,
-- Scott Smith CCNP, CDE, MCNE, MCSE
This archive was generated by hypermail 2.1.4 : Sun Sep 04 2005 - 17:01:20 GMT-3