Interface utilization numbers during heavy traffic

Steve_Pfister

We have a bandwidth utilization report that we run every month covering about 30 remote sites. This time, we are noticing that peak transmit and receive are slightly over the expected maximum bandwidth and a couple are a fair bit over (e.g., 3.17 Mbps vs. 2.23 Mbps). In starting to check this problem out, I noticed the current bandwidth utilization for the interface just mentioned was 115%. Bandwidth statements are set correctly on all interfaces.

During the month of the report in question, incoming traffic was fairly heavy. Bandwidth is nearing its maximum quite often, and a lot of errors/discards are occurring in the incoming direction. Could these errors explain why these utilization numbers are so high?

BillFitz

Hi Steve,

This is a question we see pretty regularly, and normally there are a few things to look at in answering.

First, a high number of discards means an upstream/downstream interface combination of different speeds is forcing one side to discard packets due to load at various times.  If you begin to see the error counters increase as well it means you have sufficient load that it is over-loading not only the interface logical ability to process packets, but also the physical capacity for processing data is failing resulting in errors.  While a small number of discards is pretty normal, ensuring you have sufficient bandwidth to avoid both of these conditions is important.

As for the traffic utilization, it is not uncommon for a fractional interface to burst above the committed traffic rate.  This can be seen on both fractional T1 and fractional DS3 interfaces fairly often.

Most of the time these bursts account for excess traffic being reported.

Ocassionally it is also possible that the device/interface simply hits an error in updating the counters.  This is far more likely to occur on a device experiencing high sustained loads.

An important note to remember of Orion reporting in general is that we can only deliver graphs and reports that are as accurate as the information we collect.  Unforturnately it's not uncommon that devices sometimes provide bad data.

Hope this helps.