Embedding a round-trip test in the protocol

[Jim Roskind]

One concern that I have to deal with in Chromium's experimental
implementation of SDCH is the possibility that some proxy may corrupt
a transactions. Examples may include proxies that don't understand a
content-encoding or accept-encoding header values of sdch. In the
worst case, some proxies may delete or change these headers. Deleting
a content-encoding header of sdch (when the content *is* encoded) can
be very damaging to the transmission :-/.

One interesting proposal I heard from a coworker was to turn the
dictionary download from each site into a round-trip test of the
protocol. For example, if SDCH had a specification of a standard
(preloaded but minimal) dictionary that is used to download any "site
specific" dictionaries, then the acquisition of a real dictionary
could itself be a test of the viability of the protocol across a set
of network connections.

Using this approach, if a dictionary could not be obtained, then no
further attempt to obtain SDCH encoded content from a given site could
even possibly proceed (as there is no site-specific dictionary to
advertise to the site). This in turn would probably greatly lower the
probability of using the protocol across a connection that was not
hospitable.

Comments?

Thanks,

Jim

[Wei-Hsin Lee]

Interesting idea. Please add it to the list when we are ready to review them. Thanks,

Wei-Hsin

[Lincoln]

Hi Jim:

On Oct 11, 11:51 am, Jim Roskind <j...@chromium.org> wrote:
> One interesting proposal I heard from a coworker was to turn the
> dictionary download from each site into a round-trip test of the
> protocol. For example, if SDCH had a specification of a standard
> (preloaded but minimal) dictionary that is used to download any "site
> specific" dictionaries, then the acquisition of a real dictionary
> could itself be a test of the viability of the protocol across a set
> of network connections.

The simplest definition of a "preloaded but minimal" dictionary would
be the trivial dictionary of size 0. Using this empty dictionary to
encode dictionaries for transmission would add about 20 bytes of
overhead (for the VCDIFF headers) to each dictionary file, prior to
gzip compression.

Saludos,

lincoln