Diffserv QoS Mapping.

[cedlef.net]

Hi everyone. Here is something i can't find out the answer :

I have to design a network in which LANs are working with DiffServ Classification and interworking with a MPLS backbone. The DiffServ policy is particular as there are two VoIP classes with the same low latency and low jitter guarantees and differ only on the drop precedence. In other words, when congestion occurs, mpls nodes have to drop one kind of Voice packets and continue to serve the other kind with guaranteed properties, including bandwidth, in order to maintain some of the communications.

So basically i have one VoIP queue and i need the LSR to figure out which packet can't be dropped and which can.

I thought i could use E-LSP mapping like this : two DSCP corresponding to Two VoIP classes using one LSP (when communications are going to the same Egress LSR) in which EXP is carrying out the drop precedence. Is it possible? Is there a better way?

Thanks a lot for your future answers.

Ced.

[swaroop.potdar]

Ced, you can although create your own LSP's on the physical path that you want using MPLS TE, and then also assign the traffic to those TE LSP's using the EXP ( in turn which is mapped by DSCP) using the feature Class Based Tunnel Selection.

http://www.cisco.com/en/US/products/sw/iosswrel/ps1829/products_feature_guide09186a00802659b9.html

But, I am not sure whether I would be in a position to answer whether is there a better way or not without understanding clearly what is the requirement.

HTH-Cheers,

Swaroop

[cedlef.net]

Thanks for your answer Swaroop.

Let see if i can give more details for you to understand.

The network is dealing with some common traffic like VoIP, mail, web ... The only thing is that it must guarantee the transfert of some users' communications. Users are working into an ip domain and interworking through a MPLS domain.

Into the ip domain i can configure some diffserv class marking policies onto the routers based tcp port (traffic type) and on source IP and/or destination IP and some others parameters. Say i have 5 technical classes. Each one of these 5 classes can further be subcategorized into 2 subclasses : important traffic and standard traffic. So i'm dealing with 10 dscp values that, from a technical point of view, say latency ,jitter and bandwidth, are similar two by two. In other words there is important VoiP traffic and standard VoIP traffic. Both of them need to be served with the same technical quality by each node of the MPLS domain, so belonging to the same queue. There could be also important ftp-like traffic and standard ftp-like traffic. In case of congestion, the node must preserve important traffic on each queue and only drop standard traffic. That is to say the node must continue to serve important Voip packets into the voice queue but also important ftp traffic into the ftp-like queue.

As i understand, the node can use the label (L-LSP) or its exp (E-LSP) value to select which queue has to serve the traffic. I can't define one l-lsp for each VoIP route, important and standard, and each ftp route, important and standard, into the MPLS domain for scalability problems. On the other hand, EXP only gives me 8 possible values which is unsufficient to map completely all kinds of traffic (10). So, is it possible to merge L-LSP and E-LSP : Label defining the route and also the type of traffic like Voice, ftp, ... and EXP defining whether packets can be dropped or not.

I hope i clarified my question.

Thanks.

Ced.

[swaroop.potdar]

Ced, afaik you cannot enable different drop priority based on DSCP within the MPLS core (you can but only the leftmost 3 significant bits are considered which makes it prec/EXP based) so dscp based dropping can be done only at the egress.

And also this means that you are left with 8 EXP values only for now to design you per hop QOS behaviour.

So I believe your implementation cannot be implemented practically, so you may have to consider putting the not so important voice traffic into some other class apart from the very important voice traffic.

If you are not the service provider then you will also need to talk to your service provider whether by putting your not so important voice traffic in some other class for the drop probability can he still gurantee the jitter and latency for that class which would be required for proper functioning of your not so important voice traffic.

HTH-Cheers,

Swaroop

[patrick.taylor]

Ced,

Swaroop is right. If this traverses the MPLS clould and you are using a provider (can only speak to this) you would have the provider perform the mappings between the PE and the Core ie he would take your dscp values and mark it to the EXP bits.

This again depends on how many classes of service the provider provides. Normally there are 5 classes but in your case you would have to use a voice and a video queue for your different kind of voice traffic. Your voice traffic would be marked ef (TOS 5 this would be your main) and the remainder would have to be marked to either a 4 (AF41 or corresponding dscp value) to traverse a different queue. Dont think there is any way other than this to accomplish the same.