Re: Magic Markers and MD5

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