On Mon, 7 Sep 1998, Chris Wedgwood wrote:
> Presumably, range-requests are another area where we can optimize
> thing further,
Agree. However, storing partial objects and merging parts as they come
is complicated since it goes agains the basic Squid assumption about the
object being a single continuous piece of data identified by its URL.
> possible with a big win here, but since they are
> relatively rare events and it will affect the common case code not to
> mention require slightly large swap, its probably not worth it unless
> modern browsers start placing more demands on the system.
The problem with ranges is that some applications rely on their efficient
support. We had complaints from content providers who put huge .pdf documents
on their sites, and then are surprised why it takes a long time for Acrobat
Reader plugin to access a random page withing a document.
There are a few heuristics that we can add to Squid to improve the
performance in similar scenarios, but nothing will work in 100% of cases
until storage of partial objects is implemented.
Alex.
Received on Tue Jul 29 2003 - 13:15:53 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:11:55 MST