Cell Relay Archive

[anouch_atm@yahoo.com (anouch)]

"Alvaro Osle" <alvaro.osle@cramer.com> wrote in message news:<3bdfdcb3$0$8508$ed9e5944@reading.news.pipex.net>...
> Dear all,
> 
> I want to provision ATM services between two points and I would like to know
> how much bandwidth I need to provide those services.
> 
> * I know that, in case of a CBR service, I will have to allocate the
> bandwidth equivalent to its PCR (Peak Cell Rate).
> 
> * In case of ABR, the bandwidth I will have to guarantee is the MCR (Minimum
> Cell Rate). If there's more bandwidth available in a specific moment it'll
> use that, but in order to ensure the QoS for ABR services is enough to
> guarantee its MCR.
> 
> * For UBR services, I don't need to allocate any bandwidth at all, as it is
> a "best effort" service and it'll use the capacity available in each moment.
> 
> * In case of VBR services, I can use several approaches (CAC algorithms):
> 
>    1. Peak Rate Allocation: I allocate the PCR as it was a CBR service.
> It'll always accomplish the QoS required. The problem is that no
> multiplexing gain is achieved and I'm wasting bandwidth.
> 
> THE QUESTION IS:
> 
>    2. In order to obtain statistical multiplexing gain, and hence, maximise
> the utilisation of the capacity, I know that there are thousands of CAC
> algorithms based on complex statistical models that require many parameters.
> I would like to know which of those algorithms are the most popular and the
> most widely used nowadays.
> I'd like to know the traffic parameters (descriptors) needed to estimate
> that bandwidth. I'm looking for an algorithm that requires a few traffic
> parameters (3 or 4)...
> 
> Is (in general) the Gaussian approximation an accurate CAC algorithm? Is
> there any other "better"?
> 
> Thank you very much in advanced.
> 
> Alvaro

Hi Alvaro,
I'm not sure to understand your question whether you want to write a
CAC algorithm or estimate the equivalent bandwidth required given a
service category with traffic parameters and a specific CAC algorithm.

All I can say is that Nortel Passports use the Extended Gibbens Hunt
algorithm. Regarding the parameters, any algorithm must support the
traffic parameters specified in ATM Forum TM specification. In case of
VBR, these are PCR, SCR,MBS and CDVT. It is worth noting that as other
algorithms based on statistical models, this one use the Mean Burst
Size and not the Maximum BS as specified in ATM Forum TM. The result
is that the equivalent BW calculated based on your traffic parameters
is over estimated (pesimistic). Your equivalent cell rate (ECR) should
be between PCR and SCR. Beside traffic parameters there are some other
parameters like link rate and buffer size that are taken into account
when computing ECR. For example with little buffer size your ECR
approches PCR while with bigger buffer size your ECR may be reduced
given a constant CLR.
But beside all these theoretical approches, if your objectif is to
gain on multiplexing nature of traffic you'd better to monitor
periodically the traffic. Traffic parameters are one thing, and the
real traffic pattern is another one. The best way, as far as I know to
over-provisioning is to observe the traffic and find the percentage of
the link BW left unused.

Anouch