Jing,
find my lines inserted
Heinrich
Hummel ,
> HH=> a) With respect to a particular road system X , you may determine a respective overflow road system X-overflow.
> E.g. you may determine by computation the cernel of a smallest size tree being that set of links
> which are used by at least y s/d-traffic combinations, whereby n < y < n*(n-1); y be a fixed assigned number.
> You may recompute the smallest size tree, but before, you want to mark all the cernel links to DISREGARD.
> Traffic which is about to enter the cernel of X, may be forwarded via X-overflow while changing the
> receive DSCP-x to DSCP-x-overflow. The decission to do so may be based on a currently available traffic load/
> traffic metering information.
>
what's the meaning of "cernel"? I can't find it in dictionary. But, if you means overflow those congested traffic to
another path, I think it will do harm to promising of SLA signed. Perhaps what we need is to reroute some part of
traffic to another path with no/less congestion
HH=> Sorry for my bad English: I mean kernel (not cernel).
But why would it do harm to promising of SLA ? This overflow route may provide the same QoS/SLA !
>
> b) Or you generate road systems X0, X1, X2, X3 such that the links of Xi, avoid overlaps with links of Xj, j<i.
> One micro flow uses DSCP x0, the next one DSCP x1,... as to balance the traffic and as to avoid congestion
> in the first place.
>
I don't think load balancing could be achieved by this way, as self-similar and asymetric distribution in internet
traffic
HH=> By utilizing an extra information (i.e. different DSCPs)we are much better of than ECMP, where source/dest. is hashed
in a poorly way. You really get for free the routing of all packets of a particular microflow along the same route !
>
>
>
> HH:> There is never a free lunch. But what are a view routing tables compared to the probably much bigger number
> of p2p LSPs?
>
But the problem is routers have to maintein all information about DSCP they support and propagate them.
HH=> this shouldn't be a big lunch packet, right ?
>
>
> 3. If heuritic method is not feasible, go for some approximation
> methods and prove using some theoretical results. Then support your
> performance claims with some simulation results.
>
> HH=> If there is enough interest, I would be happy to show this C++-program on work.
> As I said, it is not yet enhanced with tie breaker rules. But that would be very important as to
> determine the same identical smallest size tree determined b whichever node. If this failed we would
> have to continue with rambo-SPF:-)
>
> If you do all this, I will accept that your algorithm doesn't
> consititute any performance problem.
>
> HH=> No problem. Be sure, it doesn't run 10**80 years :-)
-- Jing ShenState Key Lab of CAD&CG ZheJiang University(YuQuan) HangZhou, Z.J. 310027 P.R.China
Tel: +86-571-87932423
Email: jshen@cad.zju.edu.cn
********************************************************************** * The SunShine of life is made up of very little beams which is * * bright all the time * **********************************************************************
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:04 EDT