Acceptable link utilization depends on the link and what type of traffic. By type of traffic I mean batch traffic (like a batch ftp process) or interactive traffic (meaning a human entering a request and waiting for a response).
A full duplex symmetrical link can get to 70-90% utilization before bad things start to happen.
However a half duplex link starts having problems at 30-40% utilization.
A full duplex asymmetrical link's utilization will vary depending on the difference between the upstream and downstream link speeds and if the majority of the traffic is flowing the the "fast" side or the "slow" side.
This is all based on interactive user traffic and how long a user will wait for a response before getting frustrated. If this is non-interactive traffic (batch ftp for example) then it really does not matter as long as the transfer finishes in an acceptable time frame.
What port? You mean the physical port, like a serial or Ethernet port? Electrons are flowing over a copper wire, or light pulses are flowing over a fiber optic cable.
I have run circuits up to 100% busy. If end users are involved, then some will complain about response time. If it is just batch traffic, it will get through. And technically circuits are either 100% busy or 0% busy. "Link utilization" is just a measurement of how often they are 100% busy. If you say your link was 50% busy over a 5 minute period this actually means that for 150 seconds the link was 100% busy and for 150 seconds the link was 0% busy.
MPLS networks are no different than any other network as far as link utilization goes. MPLS is just a different way to categorize and route traffic.
