[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: Some Comments on ID/Loc Separation Proposals



> Several folks have asked us how NOID would work with respect
> to "hosts" that appear to be a single entity yet are 
> implemented using load-balancers and multiple actual servers.
> 
> In this case, what does an "identifier" bind to?  

If the load balancing is transparent - with one IP address for
the service - then the identifier would bind to the load balancer
and the LB effectively has a private interface with actual servers.
Thus that case is out of scope.

If the load balancing is visble to the clients as multiple different
IP addresses then there is an pen issue and an opportunity that
is pointed out in the NOID draft.

In this case with NOID you can have multiple identifers=FQDNs
	www.example.com	for the service
	wwwN.example.com for each of the N servers

Since we propose adding a new "M6 capable" RR in the DNS
we have the option of making that new RR help distinguish
between case of multiple locators assigned to the same host/stack
and multiple locators assigned to different hosts/stacks providing
the service.
An example (probably not be best approach) of how this could be 
done is to have the M6 RR contain
the FQDNs for the host (the set of wwwN above) and a lookup
of the www1 etc would return the locators for that host.
Thus each host would have a unique FQDN, which would be part of 
the "cost" of using NOID together with this form of load balancing.

Other schemes might be able to use other approaches. For instance,
in CB64 all the locators for all the nodes could be in the AAAA
RRset for www.example.com and the client can see whether there
is one or multiple hosts by comparing the 64-bit hash of the 
public key; those AAAA RRs with the same hash are locators for
the same host. Hence the M6 layer can limit rehoming to be between
locators assigned to the same host.

Thus I think multihoming is an opportunity to make the
destinction between services and host names in the DNS more clear
for multihomed sites.

   Erik