Acceptable Load on WAN Link


We have a 100 Mbps Leased Line between two Cisco 7301 routers. The load on the line during the normal office hours is around 40 % but at the start of the day the load of the line climbs to a peak that sometimes goes up to 100 %, mostly it doesn't get any higher then 90 %. All traffic on the line is data ( no Voice or Video ) and the applications are non-interactive like Lotus notes mail and other database applications. Now the customer is saying that loads above 60 % are unacceptable for them, I really think it would be a waste of money to use no more then

60 % of the available bandwidth. I agree that having 100 % load is not good and was thinking on implementing RED on the routers to avoid this, but what I really need is a document that handles this topic and shows that the 60 % load is really unrealistic. Any documents with 'rules of thumb' or best practice are more then welcome, I would also like to hear your personal opinion on this.

Thanks for your feedback

Reply to
Loading thread data ...

What you are looking for is a review of queueing theory. Bottom line is that you can expect typical delay to increase dramatically as loading goes up. Assuming the delay for a single packet is X msec, the delay at 50% load will be 2 X, at 75% load it will be 4 X, at

90% load expect 10 X and at 100% load, you're approaching infinity.

The classical rule of thumb is to use a number between 60 and 70 % as the maximum design load for this reason. Good designers not only know where the numbers come from, but also know that they need to keep the numbers in perspective. If the zero load delay is barely acceptable for the application, 50% loading could be too heavy, while if the network provides 1 ms, and the application only requires 100 ms, anything less than 100% loading is fine.

Bottom line, the answer to "how much load is too much load" is the same as the answer to almost all network design questions: "It depends..."

Good luck and have fun!

Reply to
Vincent C Jones

V> What you are looking for is a review of queueing theory

Vincent knows more than I do about this however the following has occurred to me.

How about measuring the network delay directly and basing decisions on that?

e.g. use some tool to ping across the link and report the results. Cisco SAA for example.

I guess that a problem with this is that the delay measurement is likely to be a small sample of the traffic and the resulting conclusions will be subject to statistical uncertainty.

PS. Why are there no tools that sort all of this out for us?

Reply to
anybody43 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.