Basic QoS Question

[ricey]

I am setting up QoS in a VPN environemnt to prioritise voice and video traffic using dscp. I am planning to use something like the following.....

policy-map enterprise

class voice

priority 256

class video

bandwidth 512

class class-default

fair-queue

random-detect

My question relates to the priority and bandwidth statements. If no voice or video traffic is traversing the router, can other applications utilise the bandwidth reserved for these classes (i.e. does the reservation only apply when those classes need the bandwidth?)

Thanks very much

[Collin Clark]

Your thinking is correct. The only time QoS is 'in effect' is when the transmit queue is getting full. When it is not, all applications get to use all the bandwidth per your default queuing.

Hope that helps.

[ricey]

Collin,

Thats great (exactly what I was hoping to hear) Thanks very much for clarifying that.

[Joseph W. Doherty]

Some other considerations:

On many platforms, FQ flows appear to be able to compete for bandwidth against other defined non-LLQ classes. (I.e. with FQ your video class might not be truly guaranteed 512.)

When working with VPN, it's possible to have none VPN and VPN traffic share the same interface. If so, QoS configuation needs to account for that.

When working with VPN, interface QoS often "sees" all VPN traffic as a single flow.

When working with VPN, you might need to shape for far side's bandwidth.

When working with VPN, if far side shares bandwidth with other non-VPN traffic, VPN can be ineffective.

[ricey]

Joseph,

Thanks very much for the update, all very usefult tips. Re sharing with non VPN traffic, all traffic is tunneled from (even internet traffic gets tunneled to a hub site) so all traffic is VPN.

I am wondering about your comments on shaping, as the router we are using has 2 * FE interfaces and connects to an ISP router via FE but that has a serial interface with speed of 1.5 Mbps. I would like to shape all traffic at our router so that traffic is restricted to 1.5 Mbps as it leaves the FE int connecting to the ISP router. With the above example how would I ensure that all classes did not total more than 1.5 mbps? Also as an aside is there any advantage of useing Link Fragmentaion and Interleaving in this scenario?

Thanks again.

[Joseph W. Doherty]

How are you using 2 FE interfaces? Is there one link or two between you and the ISP?

How to do it, normally:

policy-map shapeForT1

class class-default

service-policy enterprise

!shaping a bit below full rate to assure congestion happens here and not at far side physical link - you can adjust

shape average 1400000

interface FE x

service-policy shapeForT1

Normal recommendation is LFI is only needed at half T1 or less.

[ricey]

Joseph,

Thanks again for your reply. When I say I have 2 FEs I mean 1 connecting to our LAN and 1 connecting to the ISP router, The ISP router then has a T1 serial adaptor so I want to shape traffic at our egress interface to (hopefully) ensure congestion does not occur at the ISP serial interface. Apologies if that was not clear.

Regarding the shaping, with the shaping applied to class-default,how would that interact with the other queues? e.g. if voice has priority of 256 and video bandwidth of 512 that would add up to more than the T1 we have. Would the above ensure that total bandwidth was restricted to 1400000 or just the class-default class? If only class-default do you know a way I can shape All classes to a combined maximum of 1400000? Sorry if I'm being a bit thick here and thanks again for your help, I really appreciate it.

[Joseph W. Doherty]

"Regarding the shaping, with the shaping applied to class-default,how would that interact with the other queues?"

policy-map shapeForT1

class class-default

service-policy enterprise

!shaping a bit below full rate to assure congestion happens here and not at far side physical link - you can adjust

shape average 1400000

[ricey]

Joseph,

So you mean the enterprise policy works out the queuing then the new policy services that and shapes there, so the policies are hierarchical and the new policy is applied to the interface. I think I've got it. Thanks very much.

[Joseph W. Doherty · Accepted solution]

Exactly!

In fact, it is called a hierarchical policy. And, you're very welcome.