Oversubscription and QOS

Unanswered Question
Jul 10th, 2010

Hi,

I am trying to understand the QOS toolsets and applying them in different scenarios.

Scenario 1:  There are multiple inputs into a switch for eg: 8 x 100Mb ports and only 1 x 1Gb uplink port out of that switch

1. Is there be any benefit in using QOS? The reason I am confused is because there is still 200Mb of unused bandwidth on the uplink there is no congestion and from reading other posts it says that QOS is only applied if there is congestion.

2. If there is no QOS than can it be that if 1 of the 100Mb input interfaces has a constant rate of traffic than would other traffic from the other 100Mb ports be subject to wait times or delay of any sort on the 1Gb uplink. What tool if any would be helpful in such a scenario?

Scenario 2: There are multiple inputs into a switch for eg: 12 x 100Mb ports and only 1 x 1Gb uplink port ouf of the switch

1. In this case I assume we would have to use QOS however which tool in the QOS toolset would be used. I am thinking only shaping. Is that correct? Would shaping result in any drop packets and would there be any other considersations?

2. is there a difference between shaping and congestion avoidance like WRED etc and queuing and how would congestion avoidance and queuing?

3. Why is it that ports have lesser receive queues and more transmit queues and why is queuing only configured in the transmit direction but not in the receive direction?

4. Shouldn't a packet be placed in a queue as soon as you receive it on the port and also to prioritize it across the backplane?

Tx

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Loading.
Jayakrishna Mada Mon, 07/12/2010 - 20:39

Hi

see answers inline:

Scenario 1:  There are multiple inputs into a switch for eg: 8 x  100Mb ports and only 1 x 1Gb uplink port out of that switch

1. Is there  be any benefit in using QOS? The reason I am confused is because there  is still 200Mb of unused bandwidth on the uplink there is no congestion  and from reading other posts it says that QOS is only applied if there  is congestion.

[JayaKrishna] Thats correct, QoS is usually applied when there is congestion and you want voice traffic to got into different Q then data traffic so that it gets prefered when there is congestion.

2.  If there is no QOS than can it be that if 1 of the 100Mb input  interfaces has a constant rate of traffic than would other traffic from  the other 100Mb ports be subject to wait times or delay of any sort on  the 1Gb uplink. What tool if any would be helpful in such a scenario?

[JayaKrishna] When QoS is turned of there is only one Q on the egress and it follows FIFO (first in first out). So there could be delays if there is voice traffic apart from this 100mb traffic from other interfaces but as I said above the interface follows FIFO so it depends.

Scenario 2:  There are multiple inputs into a switch for eg: 12 x 100Mb ports and  only 1 x 1Gb uplink port ouf of the switch

1. In this case I assume we would  have to use QOS however which tool in the QOS toolset would be used. I  am thinking only shaping. Is that correct? Would shaping result in any  drop packets and would there be any other considersations?

[JayaKrishna] Thats correct in this case you would need to enable QoS. It really depends on which platform you are using. Some switches dont support egress shaping.

you can use policing, the difference between policing and shaping is that in policing we drop packets once the CIR is increased and in shaping we buffer the packtes.

2. is there a  difference between shaping and congestion avoidance like WRED etc and  queuing and how would congestion avoidance and queuing?

[JayaKrishna] Yes, shaping buffer the packet when the CIR is exceeded. Congestion avoidance drops the packet based on the weights assinged to eliminate any connection on the interface in the egress direction.

3. Why is it  that ports have lesser receive queues and more transmit queues and why  is queuing only configured in the transmit direction but not in the  receive direction?

[JayaKrishna] We have queues in ingress direction as well but usually its egress queues that come into play when we need to prioratize traffic and its at the egress port asic where there packet is re-written if needed so that is why we have high egress buffers.

4. Shouldn't a packet be placed in a queue as soon as you  receive it on the port and also to prioritize it across the backplane?

[JayaKrishna]  its only few milli sec delay introduced and thats because of the look up done during the packet processing.

Hope that helps.

JayaKrishna

Actions

This Discussion