cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
Announcements

510
Views
0
Helpful
0
Replies
Highlighted
Beginner

Egress Queuing Assistance on Cat6509E (IOS Based)

I am looking to implement QoS on a Cat6509E running 12.2(33)SXH8a IOS.  The object of this implementation is to prioritize one specific data class while all others can be treated as default traffic.  QoS is currently disabled and thus all traffic is being handled as CoS 0.

As Switch based QoS uses hardware queues I am trying to understand the best way to implement this solution as if I were working with two queues.  A priority queue and an "everything else" queue.

After reading the documenation online I have come up with a couple of options.  I am hoping someone out there has done this before and can comment on what makes sense and what doesn't and even provide recommendations.

The line card the ports are connected to is a WS-X6748-GE-TX.

With QoS globally disabled I see the following when I issue the show queueing interface command

    WRR bandwidth ratios:  100[queue 1]   0[queue 2]   0[queue 3]
    queue-limit ratios:    100[queue 1]   0[queue 2]   0[queue 3]

queue tail-drop-thresholds
    --------------------------
    1     100[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]
    2     100[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]
    3     100[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]

queue thresh cos-map
    ---------------------------------------
    1     1      0
    1     2      1
    1     3
    1     4
    1     5
    1     6
    1     7
    1     8
    2     1      2
    2     2      3 4
    2     3
    2     4
    2     5
    2     6
    2     7
    2     8
    3     1      6 7
    3     2
    3     3
    3     4
    3     5
    3     6
    3     7
    3     8
    4     1      5

When I enabled QoS the following changes were observed:

WRR bandwidth ratios:  100[queue 1] 150[queue 2] 200[queue 3]
queue-limit ratios:     50[queue 1]  20[queue 2]  15[queue 3]

    queue tail-drop-thresholds
    --------------------------
    1     70[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]
    2     70[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]
    3     100[1] 100[2] 100[3] 100[4] 100[5] 100[6] 100[7] 100[8]


So what I can see is that the WRR bandwidth ratios changed so that Q1 now does not get 100% of the bandwidth and actually gets the lowest amount of bandwith of the 3 WRR queues.  The queue limit ratios for the buffers were also changed alloting some of the buffers to the other two WRR queues.

What I am considering doing is to:
    - Utilize the PQ and WRR Q1 and to disable the other two WRR queues
    - Assign all CoS values to Q1 except for CoS 5 which would be assigned to the PQ

Implementing the following commands on the ports
    - wrr-queue bandwidth 100 0 0
    - wrr-queue queue-limit 85 0 15
    - wrr-queue cos-map 1 1 0 1 2 3 4 6 7

I now see the following:

    WRR bandwidth ratios:  100[queue 1]   0[queue 2]   0[queue 3]
    queue-limit ratios:     85[queue 1]   0[queue 2]  15[queue 3]

queue thresh cos-map
    ---------------------------------------
    1     1      0 1 2 3 4 6 7
    1     2
    1     3
    1     4
    1     5
    1     6
    1     7
    1     8
    2     1
    2     2
    2     3
    2     4
    2     5
    2     6
    2     7
    2     8
    3     1
    3     2
    3     3
    3     4
    3     5
    3     6
    3     7
    3     8
    4     1      5

I may actually map the other CoS values to Q1T2 instead of Q1T1 so that the drop thresholds are 100% instead of 70%.

Does this make sense?  Will this give me the two queues I am looking for?  The original CoS to Queue mapping table breaks up the differenet CoS values into the different queues placing CoS 6 and CoS 7 into Q3T1.  If I lump them all into Q1T1 or Q1T2 could I be causing any issues?  Am I correct in recalling that specific core protocols might be tagged CoS 6 or CoS 7?  I am thinking it shouldnt matter as when QoS was disabled it was all being treated the same anyways (right?).

Antother thing I noticed was that when changing the CoS to Q mapping it affected a group of 12 ports.  I really only need to do egress queuing on the WAN links however the qos command affects another 11 ports around it.  Should I be concerned about this?  Would it make sense to set the CoS to Q mapping the same for all ports on the switch?

The final question I have is if I just enabled QoS globally and left the other settings as default could this have an adverse affect on the data traffic compared to when QoS was not enabled.  This would be in regards to buffers and bandwidth not being 100% assigned to a single queue.  If the only traffic in the network was either CoS 5 or CoS 0.  Could Q1 utilize the buffers and bandwidth assigned to those other two WRR queues that have no trafffic?

If so.  What would be the better solution?  Customizing it to have two queues or just having all 4 and only utilizing 2 of them?

Thanks

Everyone's tags (3)
CreatePlease to create content
Content for Community-Ad
July's Community Spotlight Awards