Re: Microsoft CARP

Hi

> I just reed the Cache Array Routing Protocol (
> http://www.microsoft.com/proxy/guide/CarpWP.asp?A=2&B=3 ) and they say
> that CARP is better than ICP because:
> 1.-ICP queries produce congestion. CARP use a deterministic request
> resolution path.
> 2.-ICP servers 'can rapidly evolve into essentially duplicates caches of
> the most frecuently requested URLs'.

> What I think is that MS does not understand quite well ICP:
> 1.- ICP are short UDP (conexionless) packets, also you can use
> multicast. In my case (I'm a ISP in Peru, and the link to USA is bad:
> 600ms minimun delay, 10% packet loss) ICP respond rapidly in changes in
> lines conditions. I dont understand quite well how deterministic request
> resolution path works, but I don't think that is as dynamic as ICP.
> 2.- If you dont want duplicates caches URLs you can use PROXY-ONLY
> parameter.

I think that the CARP idea is a good one (pity microsoft didn't
design it - see http://naragw.sharp.co.jp/sps/ :)
BTW - this includes a java script (ie an auto-config script) that allows you
to auto-balance between multiple servers.

It's main advantage is the lack of duplication - something that would
help greatly in certain cases... like ours, where we load balance
between 3 servers on the same network. I guess it's pretty useless
once you try to handoff to multiple caches where one caches US,
the other ZA (like the NLANR cache project).

Basically the advantage is that it hands off the decision of which cache
to get data from to the client (instead of the parent's talking)

If you want more technical details:
draft-vinod-icp-traffic-dist-00.txt
draft-vinod-carp-v1-01.txt

Oskar
Received on Sun Oct 12 1997 - 13:55:42 MDT