Re: Magic Markers and MD5

I missed the beginning of this... but have you looked at HTTP/1.1
chunking? It is specifically intended to send something safely for
which the content-length isn't known initially.

Oh damn, Apache 1.2 only uses chunked encoding when the client issues
a 1.1 request... I wonder if I can convince the other developers to
"extend" http/1.0 such that if the client sends "Accept-encoding: chunked"
then it will use 1.1-style chunking. chunking is a MUST in 1.1.

Dean

On Mon, 2 Jun 1997, Duane Wessels wrote:

> map@iphil.net writes:
>
> >If we use a Magic Marker sequence of bytes, how do we know that the
> >sequence will not occur in real live data? How about adding a
> >Content-Length to objects that lack one? Or, compute the MD5 checksum
> >and send a Content-MD5 header? However, we will only be able to tamper
> >with the headers on HITs; MISSes that cut through the parent proxy
> >won't have a chance. Nevertheless MD5 is useful for dupe detection,
> >which I plan to add into the push system.
> >
> >Anyone already working on this, or with some ideas? I foresee using
> >the "usual" RSA md5 routines, and with the Init, Update, and Final
> >functions called from httpReadReply. We store the intermediate context
> >in HttpStateData, and the final output in MemObject. We can also
> >put the MD5 checksum in store.log where it can be useful for some log
> >analyzers.
>
> My idea was that if the client could be identified as another Squid
> cache from the request, then the reply header would include something
> like this:
>
> X-End-Of-Transfer-Marker: XXXXX
>
> Then of course that sequence of bytes would be appended at the
> end of the object. If the client Squid didn't see the marker,
> the object would be purged.
>
> Duane W.
>
>
Received on Tue Jul 29 2003 - 13:15:41 MDT