QoS bandwidth allocation

[tabbott]

Here is a question we've kicked around a few times and still have not gotten a good answer. Perhaps someone reading this can answer it. Here is the scenerio, we have a QoS congestion management policy on our IOS router. We have a premium class, basic class and voice class. The premium class gets 30% of the bandwidth, the basic gets 10% and voice gets 150 Kbps of the 1544 Kbit line. When the circuit is not fully loaded and there are no voice calls, how is the 600 Kbits of traffic scheduled assuming 300 Kbits each in the basic and premium classes. Does each class get equal chances to be served or is it prorated based on what the relative QoS settings are even though there is not congestion and QoS is not engaged? Or is it something else. For us basic and premium are our terms and basic has traffic colored to TOS 0 while premium has TOS 3. If you have a link to the Cisco documentation on this it would be helpful to defend the answer,

One other question on this. Now assume that the link has become congested but there still are no phone calls. As we have not assigned all of the bandwidth to classes (only 40% + 150K) how is the excess bandwidth used to serve the two classes presenting traffic at a rate above that of the link? By the definition of QoS each gets their guarantee of a minimum (30% and 10%) but how is the remaining bandwidth allocated to each class?

Thank you,

Terry

[n.titchener]

Firstly, is the voice placed in a priority queue?

On any interface there is a final queue called the tx-ring, if there is no congestion (congestion is defined by the tx-ring being full) packets will not be placed in the configured queues as each time a packets is sent to an interface for tramsmission the tx-ring is first checked and if it is not full the packet is sent directly to the tx-ring and queued for tramsmission. The reason for the tx-ring is to ensure there is always traffic ready for tramsmitting out the interface. This final queue is configurable with the 'tx-ring-limit' interface command. It's reccommended that the tx-ring limit is set low when queuing is configured, this forces congestion at the interface placing packets in the configured queues.

If one of the configured queues is empty the scheduler will move to the next queue and check it for packets. This mechanism results in a proportionate sharing of any unused or unallocated bandwidth so for example if there are three queues with the bandwidth allocated 40%, 50% and 10% for each queue and there is no taffic in queue 2 (50%) queue one will effectively get 80% while queue two gets 20%, this providing the tx-ring is full.

Hope this answers your question. The Cisco QOS books cover this in geat detail.

Cheers

[timm.benkula]

This message answered a lot of questions that I had also. Thanks!

What are the implications if the voice is placed in a priority queue?

[ashok_boin]

Hi,

First of all, several people have a little misconception about QoS and their policies that BW sharing is relative according to our allocation in the configuration even without congestion.

But, the imp point to remember is, QoS and it's policies come into active ONLY when there is congestion.

Any packets which should be transmitted out will follow the path like...

Interface output queue (Software) --> TX ring (Interface)

All of our queuing techniques like priority queuing handles Interface output queues only but not TX ring.

If there is no congestion, TX ring won't be full and at that time all packets are transmitted as FIFO irrespective of your QoS design and ofcourse not necessary right.

Regards...

-Ashok.