Utilization of Ethernet links

Hello All, I am interested in any studies regarding utilization of Ethernet links, specially first level links from hosts to edge switches. To the best of my knowledge, the only studies are the 1998 and 2003 studies by Odlyzko and the 2005 study by Pang, et al presented at IMC 2005. These studies show that typical utilization over a time period of one hour or more is around 1% or less.

Are there any other studies regarding utilization of Ethernet links that you are aware of? Especially any that show higher levels of utilization than mentioned above?

Thanks!

Chamara Gunaratne

Reply to
ChamaraG
Loading thread data ...

You could start by studying your own LAN.

If you ask if someone has experienced higher utilization ? Answer yes.

Any side-effects ? Only effects you could expect from any media, peak loads may cause retransmissions and/or throttling from higher levels.

Reply to
phn

Thanks for the reply. Yes, I did measure the utilization in my LAN. And what I observed was a maximum utilization of 4% (100 Mb/s link) over 30 minutes.

There is always peak loads that last for a few seconds or minutes (ex: backups). I am more interested in utilization or network load over longer time periods like 10's of minutes.

Thanks!

phn wrote:

Reply to
ChamaraG

The most important factor would be the application environment. In a typical office-automation scenario, you will invariably see average utilization over 10s of minutes or more to be a few percent at most. Other environments can generate much greater sustained loads, e.g.: e-commerce servers, SANs, backup devices, streaming video, etc.

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

Reply to
Rich Seifert

ChamaraG wrote in part:

It is fairly easy to synthetically generate high network loads using utilities such as `ttcp`. These are really only good for testing and observing breakdown behaviour.

Rich might correct me, but AFAIK, Ethernet was never intended to run loaded. The idea was to throw massive bandwidth at the devices so simple arbitration systems like CDMSA could be used rather than token schemes.

Where users perceive "the network is slow", the usual problem _isn't_ ethernet bandwidth or utilization but:

1) Overloaded end devices (file servers) or gateways (watch for TCP capture effect on asymmetric links). 2) Broken hardware/cabling [split pairs] causing bit errors, timeouts and retransmits. 3) Broken software with excessive reliance on broadcast [NetBEUI].

Otherwise, take those moderate load studies as representative of typical practice. If you must, you could model the load variability [Poisson?] and extrapolate.

-- Robert

Reply to
Robert Redelmeier

Rich Seifert, Robert Redelmeier: Thanks for your replies.

Yes, a link connecting a server would have much greater utilization than the link connecting a desktop user. Basically, I am focused on the latter case. The context for this question is the proposal before the Ethernet Alliance to adaptively vary Ethernet link data rate in response to utilization. This is in order to reduce NIC power consumption. The white paper is available here:

formatting link
All the published studies on Ethernet link utilization and traces taken from the Ethernet LANs of several universities show that for switched Ethernet links connecting desktops to switches (which make up the vast majority of Ethernet links) link utilization is very low, usually less than 1% when measured over 10's of minutes. We have only found very few examples (heavy P2P users) where utilization is greater than 1%. And only a single instance where utilization was greater than 5%.

I would very much like to know of any large scale studies to the contrary.

-- Chamara Gunaratne

Rich Seifert wrote:

Reply to
ChamaraG

Cabling-Design.com Forums website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.