[e2e] scheduled name space
Jon.Crowcroft at cl.cam.ac.uk
Sat Apr 17 00:49:23 PDT 2004
yes, thanks - this is the point - the DNS (like other services) is a tussle space because you can do creative things with
it as per rotarie like load balancing people use - some of these involve relaxing consistncy 9or deliberately
undermining consistency) for good reasons...
so what i am doing is oiffering a VERY simple hook to control the distributed workload
of these rather than having it all centralised at the authoritative server.
sure it annoys purist keepers of the true faith, but then so do a lot of engineering optimisations...
and yes, you could probably do a lot of terrible bad woeful things, but you can always do bad things with turing
complete systems - the trick is whether the expressiveness tends to push people in the direction of positive
creastivity rather than negative (think - BGP bad, SIP not quite so bad, DNS generally ok, so this is to make DNS a
bit more expressive, but not so much so as to encourage rampant random experiments, but with some ideas of how it
might enable some cool functionality)
In missive <20040417025138.GG22362 at lcs.mit.edu>, "David G. Andersen" typed:
>>On Fri, Apr 16, 2004 at 03:29:23PM -0700, Joe Touch scribed:
>>> Jon Crowcroft wrote:
>>> >i might impleent a policy based on similar olocation (eg. using bloom
>>> >filters on the source address space) or on
>>> >request similairity (using rabin fingerprints on query keys) or whatever,
>>> >but the schedule can be different to the
>>> >result of the match...
>>> Sure - it's OK if you give different replies to different sources, but
>>> the answers have to provide (ultimately) consistent content, or you're
>>> changing what a DNS query means.
>>> Otherwise, it may be a lookup service, but it's not DNS, IMO ;-)
>> Why does DNS have to provide access to consistent content?
>>Does that mean that geotargeting through DNS responses is not a valid
>>use of the DNS system? Or only if it's not deterministic?
>> Google and others have been known to use round-robin DNS
>>to direct you to one of several geographically distributed
>>datacenters, each of which may, at various times, give
>>different replies to queries. Is this not DNS?
>>work: dga at lcs.mit.edu me: dga at pobox.com
>> MIT Laboratory for Computer Science http://www.angio.net/
More information about the end2end-interest