The MPLS WG Archive

[Shahram Davari <Shahram_Davari@pmc-sierra.com>]

Hi Daneil,

With all due respect, I don't think that your mentioned approach will work efficiently. First of all it is not reasonable to expect all routers to have exactly the same BW assignment per PHB. Secondly, it is also not reasonable to expect an E-LSP to carry traffic in proportion to their PHB assignments in the routers.

My understanding of using the E-LSP aggregate BW for admission control is as follows:

During an E-LSP setup, the aggregate BW of all supported PHBs is signaled. If all the routers in the path, have sufficient BW, the E-LSP is admitted. Now at the ingress to this E-LSP, the ingress router could do a separate admission control for each supported BA that is using the mentioned E-LSP (note this is independent of the E-LSP admission control). In other words there are two phases to the admission control, one for the whole E-LSP and another one for the traffic that is going to use that E-LSP. This is very similar to the Aggregate RSVP admission control (check ISSL WG for more information).


Regards,
-Shahram

> For an E-LSP, the aggregate bandwidth is known. This 
> aggregate bandwidth
> corresponds to the sum of bandwidths that are requested for 
> each PHB class
> that are carried inside the E-LSP. In case that the 
> distribution of the
> PHB classes is known, then admission control could be carried out.
> In networks where resources are pre-allocated per PHB class
> and where this is done consistently on all LSR, then the distribution
> of the pre-allocated resources can be used to carry out 
> admission control
> per E-LSP.
> 
> For example, consider the following distribution:
> EF gets 5%, AF1.x gets 10%, AF2.x gets 20%, AF3.x gets 20%, 
> AF4.x gets 20%;
> meaning that on a 100Mbit/s link, EF could reserve up to 5 
> Mbit/s, AF1.x up
> to 10Mbit/s, etc. etc.
> 
> If now an E-LSP requests 7 Mbit/s and this E-LSP carries EF, 
> AF1.x and AF3.x
> traffic, then the 7 Mbit/s could be split up and assigned to 
> the classes
> as follows:
> EF - 1 Mbit/s
> AF1.x - 2 Mbit/s
> AF3.x - 4 Mbit/s
> according to the 'weights' of the resource distribution.
> Then, admission control for each class could be done 
> seperately. The E-LSP
> could be admitted if all of the admission control checks succeed.
> 
> Does this sound reasonable?
> 
> Of course, the whole scheme does not work if the 
> pre-allocated resource distribution
> is not consistent amont all LSR routers.
> 
> Best regards,
> 
> -Daniel
> 
> 
> 
> Marcus Brunner wrote:
> > 
> > Shahram,
> > 
> > In many scenarios, the admission control for a PBH 
> scheduling classes is
> > already fuzzy, which means on a statistical level, you will run into
> > trouble if you want to do admission control for aggregates 
> of different
> > PHB classes.
> > 
> > Marcus
> > 
> > Shahram Davari wrote:
> > >
> > > Hi Daniel,
> > >
> > > You can do admission control for E-LSP too. But the 
> admission control is done for the aggregate of all supported 
> PHBs in that LSP.
> > >
> > > Regards,
> > > -Shahram
>