Re: [TLS] Wrapping up cached info

[Nicolas Williams <[e-mail]>]

On Wed, May 19, 2010 at 01:43:52PM -0400, Kemp, David P. wrote:
> This will work as long as there is high "cache reuse ratio", i.e.,
> repeated handshakes with a server will not result in many cached objects
> of any type.

I think this is very likely.

> If multiple handshakes with a server could cache different objects of a
> given type, then the list in step 2 (client sends hash value of every
> object it has collected) could get very large and a more traditional
> cache miss approach would be needed.

I think this is unlikely.  I'm not sure how one might design an
extension to provide for caching that can handle this possibility
without adding round-trips to the handshake and without using HTTP (or
some such protocol) to fetch cacheable items.  The no-extra-round-trips
constraint is almost certainly non-negotiable.  Downloading cacheable
items out of band would greatly complicate implementations and would
introduce more failure points.

You might demand real-world data on cacheability and predictability of
caching opportunities.  But I think it should be fairly obvious that: a)
the items in question change infrequently, b) the handshakes between any
one {client, server} pair will generally all look the same, with very
few variations (initial vs. session resumption, outer handshake vs.
inner re-negotiation).  Moreover, clients could observe that use of this
extension provides no benefit and cache that to avoid using it
subsequently (much like rsync, the client could compute a "speedup"
ratio of data that would have been transfered to data actually
transfered).

Nico
--