[ietf-dkim] 1193 considered harmful

[Douglas Otis]

On Mar 21, 2006, at 3:36 PM, Michael Thomas wrote:
>
> But you'd assumedly need a new RSA signature per message in order
> for this to make any sense. To be pedantic:
>
> CostRSAsign = 1
> CostSHAx    = .1
>
> For message one, cost = 1.1, for message 2-n, cost = 1.0/msg -- big  
> whoop. These aren't accurate, just illustrative.

For SHA-1, a rough estimate of the overhead seems to suggest the hash  
is predominate at about 100KB messages.  With SHA-256 this crossover  
might become 50KB.  It does suggest that distributing larger messages  
will increase overhead by a factor of recipients without a means to  
mitigate the HASH algorithm for larger messages.

In terms of adding a new header at subsequent stages, this parameter  
remaining unchanged in a new signature to encompass the new header  
also shows that the body of the message is not being changed.  If one  
signature fails due to the HASH not matching, it can be assumed all  
signatures will fail due to a change that has been made in the  
message body.  This change may help facilitate processing signatures  
that wish to securely add information to the message.  Rather than  
causing a cascade of failures being processed, this effort can be  
quickly short-circuited.  When the HASH in each signature are  
different, then looking for a signature with a matching HASH also  
short-circuits the discovery effort which also reduces the work  
caused when multiple signatures are allowed to exist within a message.

-Doug