Catalyst QoS/Buffers/queueing on Isolated Network

[ds6123]

Hello,

Suppose I have a completely isolated network that only has a bunch of cameras sending video to a bunch of servers that are responsible for recording the video.  By default, the cameras just leave the DSCP/COS bits to 0/0 and the switches don't have qos enabled.

I'm trying to figure out, in order to get the best performance (as little packet-loss/jitter),  if I should:

#1  Leave all the default settings (No markings being done on the cameras and no queueing at the switches).  Since the purpose of qos it to ensure certain classes of traffic are allocated a certain amount of bandwidth, and since all traffic belongs to a single class, then qos is pointless.

#2  Mark the video with a DSCP/COS of 46/5 and make sure all the traffic is mapped into the priority queue on the switches.

Since this is a 100% isolated network, it seems option #1 is clearly the winner.  But digging into this deeper, I have a few questions.

A.  If I did use option #2, then I suspect eventually traffic would be policed (once it reaches 1/25th of the egress bandwidth if using auto qos) since the priority queue is shaped (even if the other 3 queues are idle).

B.  In option #1, then I'm assuming 100% of the egress buffers are allocated to the traffic, correct.

C.  What if one of the switches *did* have QOS configured, but the cameras didn't mark anything (defaulted to DSCP/COS of 0/0).  So *all* of the traffic would be mapped into the default queue.  Since the other 3 queues (two of which are shared at 25% and the priority queue is policed at 4%) are idle, would traffic be able to overflow into the other queues?  IE, would it:

     c1.  overflow into the other 2 shared queues (for a grand total of 75% of the egress bandwidth)

     c2.  overflow into the other 2 shared queues plus the shaped queue (for a grand total of 79% of the egress bandwidth)

     c3.  overflow into the other 2 shared queues plus the shaped queue (for a grand total of 100% of the egress bandwidth)

     c4.  not overflow at all so you're limited to 25% even though the other 3 queues are idle

I'm getting conflicting information.  TAC is saying it's c4.  I'm hoping/thinking it's at least c1 but possibly c3.

Thoughts?

[ds6123]

I found this awesome article so, in this case, the answer appears to be c3 -- "overflow into the other 2 shared queues plus the shaped queue (for a grand total of 100% of the egress bandwidth)"

[Joseph W. Doherty · Accepted solution]

Disclaimer

The  Author of this posting offers the information contained within this  posting without consideration and with the reader's understanding that  there's no implied or expressed suitability or fitness for any purpose.  Information provided is for informational purposes only and should not  be construed as rendering professional advice of any kind. Usage of this  posting's information is solely at reader's own risk.

Liability Disclaimer

In  no event shall Author be liable for any damages whatsoever (including,  without limitation, damages for loss of use, data or profit) arising out  of the use or inability to use the posting's information even if Author  has been advised of the possibility of such damage.

Posting

On a switch, with all homogeneous traffic, you're correct, there's little advantage to using any QoS.

For A if egress queue one is activated as an expedited queue, there is no implict policing nor bandwidth limit.  Frames in this queue will be forwarded first, using up to 100% bandwidth.  If expedited queuing is not activate, queue one, by default, is shaped at 4% bandwidth, not policed.  (NB: shaping will attempt to buffer an over bandwidth frame, policing normally just immediately discards.)

For B, yes an no.  Buffers are for traffic, reservations differ when QoS is activated.  Not all buffers, are all available, since some are dedicated to the port or port/queue (at least by default).

For C, a shared queue can use bandwidth not otherwise being used.  Its queued frames don't overflow into other queues.  I.e. it could use 100% of the bandwidth.

Cisco paper on 3750 queuing: http://www.cisco.com/en/US/products/hw/switches/ps5023/products_tech_note09186a0080883f9e.shtml#concept31

[ds6123]

Thanks for the response, Joseph.  I learned something today.