Re: [PCN] CLEBAC vs. OBAC

[Michael Menth <[e-mail]>]

Hi Tom,


Tom Taylor wrote:
> Michael, in your presentation you distinguished between Congestion 
> Level Based Admission Control (CLEBAC) and Observation Based Admission 
> Control (OBAC). CLEBAC uses a ratio estimate of congestion level, 
> while OBAC is based on observation of one or more marked packets in an 
> interval.
You refer to http://www.ietf.org/proceedings/08mar/slides/pcn-8.pdf
>
> Looking at your charts, I note that overall, OBAC provides a higher 
> probability at any level that a given flow will be blocked (not 
> admitted). 

Yes, that must be the case.

> Looking at the desirable effects, we see that although CLEBAC results 
> in distinctly less unnecessary blocking at lower traffic levels, OBAC 
> results in less over-admission at high traffic levels. Do you think 
> the latter difference is significant? 

It can be significant when PCN is used for what it is not made for: low 
ingress-egress aggregation. I don't say that this is a corner case or 
not interesting, but it is a case where PCN has problems. If you want to 
go to as small ingress-egress aggregates as possible, then OBAC might be 
a solution for such scenarios to mitigate the effect of false admission, 
but it cannot solve them.

> If so, we may want to define regimes where CLEBAC respectively OBAC 
> operate.

I don't think that it is possible or desirable to have both CLEBAC and 
OBAC to control the same ingress-egress aggregate. The large number of 
false negatives for OBAC disappear as soon as the token bucket size is 
set to larger values (see attached file), but it also delays the 
reaction of OBAC since it takes longer until marked feedback arrives 
when the admissible rate on a link is exceeded. Well, those are tradeoffs.

Just to make it clear: I think that the major benefit of OBAC compared 
to CLEBAC is its fast reaction time which may be helpful in case of 
flashcrowds where you need a fast reaction and don't want to wait 1-2 
measurement intervals which is needed in case of CLEBAC.

Regards,

    Michael Menth

>
> Tom Taylor

-- 
Dr. Michael Menth, Assistant Professor
University of Wuerzburg, Institute of Computer Science
Am Hubland, D-97074 Wuerzburg, Germany, room B206
phone: [phone], fax: [phone]
mailto:[e-mail]
http://www3.informatik.uni-wuerzburg.de/research/ngn

_______________________________________________
PCN mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/pcn