Traffic prioritization

[suthomas1]

Hi, we have two sites which are connected via MPLS.MPLS provider device is connected at both ends to an L3 switch.

If we need to prioritze or assign more bandwidth to a particular application flow based on destination address, where should the QOS be ideally configured..on the L3 switch at the end traffic is emanating from or on the MPLS device?

Small config would also help.

Thanks !

[suthomas1]

That gave a good understanding on the topic.Thanks for getting so detailed.

Voice we know is very much sensitive to delay..would this apply also to applications which hit the server and try navigating/retrieving through them.

thanks!

[Joseph W. Doherty]

It can impact other applications, again a question of degree. As you've noted, something like VoIP is about the most sensitive. Next might be remote desktop applications (driving the screen), next then might be other interactive applications. With VoIP, fractional second delays could be an issue, with interactive applications occasional multi-second delays might be just fine. (Of course, much also depends what you're willing to tolerate or consider "normal". Many, many years ago I used a system that offen had a 30 second delay to scroll a line; not that I was happy using it, but that's what was provided.

[suthomas1]

Thanks!lets say there are 2 ends A,B.Router at A has interface fa0/0 linked to router B on interface fa0/1.But fa0/1 has been subinterfaced for two seperate links fa0/1.101 , fa0/1.102 landing in.

So,link of interest here is fa0/0 to fa0/1.102.

In this case if i need to apply service policy with minimum bandwidth guarantee features ,would fa0/1.102 on B accept the service policy OR wat is the best way to do this.

Thanks in advance..

[Joseph W. Doherty]

Don't recall I've tried it with Ethernet subinterfaces, but what you would want would be either a QoS policy on the parent interface, if the physical port is the congestion point, or on the subinterface if you need to shape the subinterface bandwidth to match the downsteam bandwidth. You would also want to insure either all subinterfaces are shaped and their sums don't over allocate the port's bandwidth, or have QoS policies at both the parent and subinterface levels.