Possible QoS issue

[enriquebs]

Hi,

This is the picture: CE -- PE. Voice and Data comms are taking place here. The QoS appears to be working propperly but when there is a big load of traffic the ATM interface of the CE starts lossing packets, quality of the Voice calls gets degrade. At the same time the CPU goes up until 90% or so... QoS have been reviewed, hardware changed, etc... We're thinking that maybe the big amount of traffic makes the CPU goes up, then the router starts discarting traffic ramdonly without looking at priorities... Could be a solution creating a control-plane service policy in order to avoid the CPU increase? What do you think?

Regards,

[aravindhs]

Hi,

What router is it ? In general, the amount of traffic traversing the router interfaces should not matter as long as the switching path is NOT process based (through the CPU).

Have you checked which process takes up the CPU during high load ? That would be the first thing to look at.

Cheers

Arav

[Kevin Dorrell]

Are these outgoing packets of incoming packets that it is losing? If it is outgoing, then that may be a quite normal tail-drop according to your QoS policies. If it is incoming, then it is more likely to be related to your CPU problem.

Are you doing anything that would take a lot of CPU. For example, are you doing real-time data compression in software - that can eat your CPU in no time. Could you find out what it is doing with the 90% CPU.

Kevin Dorrell

Luxembourg

[marikakis]

Hello,

What is the maximum speed of the interconnection (I mean your access rate)?

How much exactly is the "big load of traffic" that you mention and how do you measure it?

What is the CPU usage you normally see on this router? What kind of router model have you got?

Kind Regards,

M.

[Joseph W. Doherty]

It would help if you could post parts of your config showing your QoS policy and the ATM interface.

Something to note, especially with ATM, is the possible need to override the physical FIFO interface queue to spill into your software queues sooner.

[enriquebs]

Thanks a lot for the replies, the issue was located in the ATM network.