ietf-mailsig
[Top] [All Lists]

RE: In response to Housley-mass-sec-review

2005-03-06 10:22:39

That works for me, an arbitrary opaque segmented string, allow the sender to
segment their namespace any way that makes sense to them. 

This would also address the question on partitioning that was raised. Yahoo
with its 100M users might have their names segmented into 1000 partitions
and then map those onto 20 servers as needed for load balancing purposes. It
is also possible to get RAID style overlapping redundancy without excessive
cost.

Server A Serves partitions 1-50 

Server B backs up partitions 1-3
Server C backs up partitions 4-6
...

This way the loss of one server does not cause a massive increase in load
for another server.


-----Original Message-----
From: Andrew Newton [mailto:andy(_at_)hxr(_dot_)us] 
Sent: Sunday, March 06, 2005 11:19 AM
To: Hallam-Baker, Phillip
Cc: mlibbeymail-mailsig(_at_)yahoo(_dot_)com; 'Douglas Otis'; MASS WG
Subject: Re: In response to Housley-mass-sec-review



On Mar 6, 2005, at 10:57 AM, Hallam-Baker, Phillip wrote:

Nah, why not just do a hierarchical query?

Base64(SHA1(messageID)).Base64(sha1(userID))._revocation.example.com


Standard DNS config can then be used to revoke the user or the
individual
message:

*.Base64(sha1(userID))._revocation.example.com TXT "status=revoked 
reason=spam"

I was thinking the same thing.  Though it need not be codified in the 
standard.  The identifier just needs to conform to DNS label 
syntax, be 
it one label or many.  That way Yahoo can use this type of scheme and 
smaller mail systems can use simpler schemes.

-andy




<Prev in Thread] Current Thread [Next in Thread>