On 18/04/2013 1:00 a.m., babajaga wrote:
Hi,
is it possible to re-organize all the posts regarding this topic ?
I am asking, because of special interest, and the posts regarding this issue
are spread all over different threads, unfortunately.
Possibly. Although I'm not game enough to edit the headers of each
thread message to link them.
Now my few bits: When squid caches up to several versions of the same URL,
depending on encoding (gzip, deflate, etc.) then the question arises: Is'nt
it reasonable to store just one variant (unencoded OR gzip), and do a
conversion on-the-fly, in case another version requested ?
Not all encodings are translatable. For example sdch used by Chrome is
highly compact but is effectively a diff patch against the client cached
object.
Then again encoding is only one header. Any header the server wishes to
look at can be listed in Vary:, catering for all of them in every
potential permutation is a huge consumer of CPU cycles as Anita is about
to find out when this prefetcher hits normal traffic.
Would not only help the prefetching, but for actual use, as well, to save
some cache space.
Or am I wrong here ?
Squid is slow enough as it is without adding this to the processing
overheads. We trade plentiful disk space for scarce CPU cycles.
There is a Key: header being proposed to extend Vary: header in a few
nice ways which enable more efficient caching of variants.
http://tools.ietf.org/html/draft-fielding-http-key-02
Amos
