Re: [Monotone-devel] Question on layering

[Daniel Carosone]

On Thu, Feb 22, 2007 at 04:47:40PM +0000, Paul Crowley wrote:
> Ethan Blanton wrote:
> >There are no MAC "packets"; there is a MAC appended to every
> >higher-layer netsync object.  For small objects, that would be
> >nontrivial overhead.
> 
> That's what I meant.  So a significant proportion of the stream is MAC; 
> SSL would reduce our bandwidth demands noticeably.  That sounds good.

No. The MAC might be significant for small packets (eg, refinement)
but in the cases where there's large network traffic at all, it's
because there's large content (data and deltas) to send; then the MAC
overhead is much less significant.

The overhead of SSL might even be greater, if you had large files and
large deltas being split into lots of SSL stream records.

It's all beside the point. For the original context, the comment about
network traffic for a full pull vs partial is not about overhead, but
about raw data - the point being that even the data was undesirble,
regardless of the overhead.

> SSL would authenticate the stream of bytes directly before it was
> broken up into packets, but the effect is exactly the same, isn't
> it?

If SSL was inadequate for our requirements in this regard, lots of
other people would have far bigger problems, certainly.

--
Dan.

pgpIsdZjHnuKz.pgp
Description: PGP signature