Quality of service LAN for 2 kind of traffic

[kruko90]

Hi all,

I am trying to find out what the best way is to implement quality of service within the LAN for 2 kind of traffic (VoIP and OAM). all other data is best-effort.I am using a Cisco catalyst 2960x and 3850.

Since the 2960X has 4 egress queues. I was thinking to divide the data like this:

Q1: PQ for voip (signaling and rtp streams)

Q2: OAM

Q3: best effort data

Q4 : ?

do I need to use the 4th queues for defining the 3 kinds traffic? and I looked up for best practices for a 3 class model but couldn't find anything unless the 4,8 and 12 class model. Are there any best-practices that defines the bandwith allocation to the egress queues as well? a guide by Tim Szgeti (http://d2zmdbbm9feqrf.cloudfront.net/2015/usa/pdf/BRKCRS-2501.pdf) says to use  a srr-queue share of 1, 30, 35 and 5 where the 1 is ignored cause the use of PQ. 30% for Q2 and 35 for Q3. Q4 is scavenger with a allocation of 5%. does these values needs to be 100% together?

[Joseph W. Doherty]

What the correct egress bandwidth settings should be depends on the service needs of your traffic.

On most switches, PQ places all other class traffic at risk of total bandwidth starvation.  So, ideally, PQ traffic admission should insure the traffic submitted to that class is valid, and also ideally, such traffic shouldn't usually require more than about a third of the available bandwidth.

BE traffic should probably be insured some bandwidth, say 1%.

What your OAM traffic allocation should be, as noted above, depends on the traffic's service needs.  Minimally, you'll want an allocation at least a little larger than OAM's average bandwidth requirement but it also might need to be much larger (2 or 3x as much) if you're trying to avoid queuing latency.

There's no requirement to use 4 queues if you don't want or need to.  However, you might want to use a 4 queue model that you can use for generic purposes.  As to whether the percentages need to equal 100%, that may depend on the platform.

BTW, a 4 queue model is often use is:

class real-time
 !PQ or LQ but not to exceed about a third of bandwidth
class foreground
 bandwidth remaining 81 percent
 !e.g. transactional
class background
 bandwidth remaining 1 percent
 !e.g. bulk data transfers
class class-default
 bandwidth remaining 9 percent
 !e.g. BE traffic

[kruko90]

Hi Joseph,

Thanks for the answer. You say that i isnt required to use 4 queues? but in in the configuration guide i see that there are always 4 queues configured. i look the platform 2960x if the total amount needs to be 100% but i couldn't find that out.

and is the assingment of traffic in those queues enough to prioritize it? by that i mean if i assing OAM to Q2 and BE to Q3. and lets say i use 30% for Q2 and 35 for Q3. How does the switch know how to prioritize Q2 before Q3. OAM traffic is tagged with CS2.

[Joseph W. Doherty]

Yes, there's are four hardware queues, but you don't have to direct traffic to all of them.

If you assign 30% to Q2 and 35% to Q3, the scheduler attempts to insure those two queues get that ratio of bandwidths.

Strictly speaking, only Cisco's (very old) PQ prioritized between queues.  When you're dealing with ratios, you're dealing with (sort of) relative prioritizes.  For example, if Qx was assigned 99% and Qy 1%, and both queues had traffic to transmit, most of the time Qx would be dequeued before Qy, so it behaves like it has priority, but actually it does not because Qy could be dequeued before Qx to obtain its 1%.

Back to your question, with 30% assigned to Q2 and 35% assigned to Q3, Q3 would behave as if it had a slight priority over Q2.

As to how a switch knows to treat traffic, it treats traffic based on what queue it's within and how a queue is allocated bandwidth.  As to how it knows what queue to direct traffic to, that too depends on device configuration (often there's a default).