[c-nsp] Monitoring congested links

[MKS]

Hello list.

The company I work for has some GigE circuits that are pretty long and
therefore expensive, so it's important to get the most out of them.
We are monitoring the load via basic tools like mrtg and  rtt/drop/jitter
via cisco jitter probes (ip sla/rtr).. We have qos configured on the links
because some of the traffic is sensitive (traffic groups: voip, multicast,
business and internet)

I'm curious to know what is your preferred way of monitoring (congested)
links? The mrtg graphs give 5min average traffic, so there must be
considerable bursts. When do you consider the link to be full? Then the 5min
average load reaches 50%, 60%, 70%, 80% or 90% load.

Or is it maybe better to monitor the link by using drop statistics, via
jitter probes or interface drop/qos counters?
When do you consider the link to be full? Then the average drop percent
reaches 0.1%, 0.5%, 1%, 2%, 5% for best effort traffic? (internet)

Can someone suggest some tool(s) to graph the drop/qos interface counters?

Any comments will be great.

Regards
MKS