On Fri, 19 Aug 2005, Jeffrey Haas wrote:
> On Thu, Aug 18, 2005 at 03:57:23PM -0400, Dean Anderson wrote:
> > How do they do this without a flow cache?
>
> In an attempt to try to shorten this particular part of the thread:
>
> Given a prefix, P with a set of nexthops nh1-nhx, forward
> traffic to a destination p in P by calculating a hash over some
> data in the header that maps deterministically to a given nexthop bucket.
This would seem to fall under "some similar mechanism" that I was positing
in the paragraph above the one your replied to. I should have added the
same qualification to my question, too. My point is that these mechanisms
are not required for an Internet Gateway and that PPLB is allowed
behavior, whatever your opinion of its performance implications.
Lets just summarize what we agree on:
No one seems to be saying that this anycast extension will work when PPLB
is used. Lets forget about whether PPLB is a good idea or how one might
implement course-grained load balancing.
Is there a consensus that this anycast extension will not work with PPLB?
Do we agree that a clear warning about the incompatibility of this anycast
extension with PPLB should be prominently placed in the draft?
Do we agree that for this anycast extension to work properly, one must
guarantee that 2 or more successive packets must go to the same Anycast
host?
--Dean
--
Av8 Internet Prepared to pay a premium for better service?
www.av8.net faster, more reliable, better service
617 344 9000
_________________________________________________________________
web user interface: http://darkwing.uoregon.edu/~llynch/grow.html
web archive: http://darkwing.uoregon.edu/~llynch/grow/
