In message <[EMAIL PROTECTED]>
Acee Lindem writes:
>  
> Hi Yue,
>  
> I can't explain any better than Paul and Curtis already have in their
> postings.  The problem you're seeing is either in you're OSPFv3
> implementation or in the anycast address discovery and advertisement
> in the OSPFv3 router originating the intra-area-prefix-LSA. Note that
> the latter is outside the scope of the OSPFv3 protocol.
>  
> Thanks,
> Acee


Hi Acee,

Yue has shown remarkable good character in readily accepting that
OSPFv3 may already accomodate the needs that he was addressing in his
draft.  I look forward to future contributions from him.

It does sound like he has tried to make use of anycast and run into a
problem of some sort.  Having myself only worked with IPv4 and OSPFv2
I hope that I have not missed any of the details of IPv6 address
discovery that may be at issue or some other IPv6 specific detail.

I've used a form of anycast with IPv4 and it was not without problems.
One problem we ran into was with the use of TCP and ECMP.  Routers
outside of our control did per packet BGP ECMP (rare) or periodically
reseeded the BGP ECMP hash used for src/dst split (common).  Using an
unmodified TCP application didn't work well.  Some means of
redirecting to another address on the box was needed when the initial
request was processed to allow state to be retained and "latch in" to
the same server for the duration of a session requiring state.  For
example, for HTTP today a simple redirect to another address on the
server was adequate.  Redirecting other applications would not be as
simple.  This was at a time when DNS servers did not round robin the
A-records in responses and hosts always went to the first listed.  As
I remember, it also predated widespread browser support for a fully
transparent HTTP redirect.  Eventually other techniques were used
rather than this form of anycast.  This was for a major web site (not
AOL) on a major provider (yes, it was ANS), so it was a rather large
scale early real world use, or attempted use, of anycast.

My instinct is that the same should work for IPv6 just as well (or
poorly) but I don't have any experience to back that up.

Curtis

_______________________________________________
OSPF mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/ospf

Reply via email to