QoS and routers

[Vinny]

Hi,

I have a client that have VOIP on several sites over a MPLS network. In their branch offices, they have cisco 2911/2951 with some QoS configuration.

They talk to phone servers over the MPLS to a data center that have ASR 1000 routers at edge. Problem is, these ASR 1000 routers don't have any QoS config.

I know that routers don't remove the dscp tag but are they really using it if there is no qos config ?

I was under the impression that all devices along the path should have QoS config with no exception of these ASR 1000. Note that their provider do QoS on the mpls network.

What do you think ?

sorry for my english

[Joseph W. Doherty · Accepted solution]

I know that routers don't remove the dscp tag but are they really using it if there is no qos config ?

Correct, by default routers don't change the ToS, but also by default, they generally don't do any QoS (unless configured to do so).

I was under the impression that all devices along the path should have QoS config with no exception of these ASR 1000

QoS is generally only needed where there's congestion adverse to the service needs of the application, and where QoS can mitigate.  However, as adverse congestion can "sneak up" on you, is one reason the "book" advocates it be configured end-to-end.

Note that their provider do QoS on the mpls network.

It's been a while since I've worked with MPLS service providers' QoS, but when I did, often it was a "weak" implementation (often sort of the QoS support you might find on a switch).

[Vinny]

thanks for this complete anwser !

[Vinny]

Hi have some more questions about QoS.

Let's say I have an output policy-map that looks like this :

    
policy-map Output-Policy
    class Output-Media-VOIP
        priority
    class Output-Signal-VOIP
        bandwidth remaining percent 5
    class Output-Control-Mgmt
        bandwidth remaining percent 5
    class class-default

1- As you can see, voice data is set as priority (30% default I think?). Does that mean that 30 % of the bandwidth is ALWAYS reserved for voice ? When there is no congestion and voice data use, let's say 5 %, can other queues use the bandwidth left or priority queue is fixed and cannot be shared ?

2- Does the order is important. In that example, Signaling is above Control protocols. Does this makes signal more important ? If I have ato add new classes, should I review my order before adding them or I can just add them as I want ?

thanks !

[Joseph W. Doherty · Accepted solution]

Not sure what the priority bandwidth allocation is when not specified.

#1 As a general rule, bandwidth reservation guarantee that bandwidth to a class, but do not hold it when it's not used.  Priority class bandwidth settings also generally set an implicit policer that can limit that class to the amount of bandwidth specified, but again, unused bandwidth is available to other classes.

#2 No, placing one class above another doesn't make it more important, but classes are matched in the sequence listed.  I.e. if more than one class matches a particular kind of traffic, the traffic will use the first class it matches.

[Vinny]

I have another question (I hope it's the last one!)

I'm used to apply qos based on percentage (I only did LAN qos, not WAN).

Now I have WAN interface that connect at 100mbps or 1gbps but the actual link speed is 5 or 10 mbps. If I use percentage, I'm sure it will not work as expected.

There is no bandwidth command set on WAN interfaces of these routers yet. If I set it, will QoS use that value to calculte percentage ?

thanks

[Joseph W. Doherty]

I have another question (I hope it's the last one!)

Not a problem - asking questions is the whole point - this doesn't have to be the last.

Now I have WAN interface that connect at 100mbps or 1gbps but the actual link speed is 5 or 10 mbps. If I use percentage, I'm sure it will not work as expected.

There is no bandwidth command set on WAN interfaces of these routers yet. If I set it, will QoS use that value to calculte percentage ?

I believe QoS does use it.  However, when you have available bandwidth less than the port's bandwidth, for QoS to be effective, you need to have a parent shaper.  Otherwise, QoS dequeuing prioritization won't engage until there's port congestion.

Also with percentages (or bandwidths in bps), what's really happening is you're setting relative dequeuing weights.  Basically, you get the same results if you set class to 1% and another to 2%, as if they were 5% and 10% or 30% and 60% or 3 Mbps to 6 Mbps.  I.e. anything that makes the ratio 1:2.

[Vinny]

So, in my case, would you recommand doing percentage or setting bandwidth in bps ?

If I calculate the amount of bandwidth needed depending on the number of employees in each branch offices then set class map like :

X bps for voice media

X bps for signaling

1.5 mbps for video (there is only one video system in each site with 1.5mbps of maximum bandwidth)

Is that a good practice ?

If percentage is best and I need to do a parent shaper, I suppose I don't need the bandwidth command on the interface in that case ?

thanks again

[Joseph W. Doherty]

I normally use percentages.  That way, you can use the same policy on different bandwidth links, and I think in ratios more often than actual bandwidth allocations.

For example, my generic all purpose CBWFQ policy is:

policy-map Sample
 class real-time
  priority percentage 33
 class foreground
  bandwidth remaining percentage 81
  fair-queue
 class background
  bandwidth remaining percentage 1
  fair-queue
 class class-default
  bandwidth remaining percentage 9
  fair-queue

I would still set the bandwidth, to reflect what's available, as it also provides that information for bandwidth measurement tools and some IGP routing metrics.

[Vinny]

thank you so much

So with the config above, you attached it to a parent shaper. Is that it ?

So it would look like :

policy-map Sample
 class real-time
  priority percentage 33
 class foreground
  bandwidth remaining percentage 81
  fair-queue
 class background
  bandwidth remaining percentage 1
  fair-queue
 class class-default
  bandwidth remaining percentage 9
  fair-queue

!

policy-map SHAPE

     class class-default

          shape average 10000

          service-policy Sample

interface FastEthernet0/0

   service-policy output SHAPE

[Joseph W. Doherty]

Correct.  Also, BTW, I believe most shapers count L3 throughput, but CIR limits are usually L2.  So, I've found setting the shaper to about 15% slower then the CIR rate allows for average L2 overhead.

[Vinny]

I'm reading about the fair-queue command. I never used it before. I understand the queues are serviced in Round-robin manner instead of FIFO.

I'm just not sure what is the impact of this

[Joseph W. Doherty]

The impact of FQ, a "packet train" from a bandwidth hog will not be all dequeued before a single packet from a light consumption flow.  I.e. The light consumption flow will be interleaved with the bandwidth hog's packets.

Additionally, the bandwidth hog is likely to have tail drops, from its flow queue, before the light consumption flow does (perhaps also signalling the bandwidth hog to slow its transmission rate).

[Vinny]

Thank you very much again. I really appreciate it.

Other thing. Let's summarize. I have several branch offices with 40-50 employees each. Speed of each WAN links isn't the same eveywhere BUT the number of employees is always the same. I don't want to apply different WAN policies depending of their link. So percentage would be the best I think but I suppose some sites will have too much reserved. I just need to find a percentage that even on the smallest link, is enough for the voice and signaling.

Now, some sites (about 10 out of 120) will have a video confering system. Only one per site and there are all configured to max 1.5 mbps. Because there is only on per site no matter how many people work there, it will always be the same value of 1.5mbps.

Problem is, you cannot create policy-map with percentage and a static bps settings.

So I need to create a policy that can guarantee VOIP for 40 employees and a 1.5mbps video system no matter what the wan speed is.

I have to admit I'm a little lost finding the right values for each class..

And I'm thinking about creating a policy without video and one with it depending of the site. Don't know if it's a good idea.

[Joseph W. Doherty]

So percentage would be the best I think but I suppose some sites will have too much reserved.

Remember, bandwidth isn't reserved it's only guaranteed.  I.e. set the bandwidth value too high doesn't deprive other class of bandwidth unless the class with the too high setting actually want to use the additional bandwidth.

So I need to create a policy that can guarantee VOIP for 40 employees and a 1.5mbps video system no matter what the wan speed is.

Well that's perfect then.  I.e. if you set a percentage too high for video, it will not use the "extra" guaranteed bandwidth.

I have to admit I'm a little lost finding the right values for each class..

If you were to use my generic policy, place VoIP in real-time and video in foreground.