Re: thoughts on memory usage...

--MimeMultipartBoundary
Content-type: text/plain; charset=us-ascii
Content-transfer-encoding: 7bit

Hello!

While we are on this subject...

There are some thoughts on how squid should alocate memory.

Problem with curent (1.1) alocation is that in transit files must
be small (4 MB default), or else memory allocation grows rapidly
and squid eventualy starts swaping.

NOVM version is better. It can use large files (latest programs
distributed over net can be 10 - 20 Mb ). But must read every file
from disks. This is not so good for small files and for fd usage.

If we would combine standard and NOVM version we could set some
size over which squid would go from standard (get in memory,
then swap to disk) to NOVM (swap directly to disk). This
seting would be user definable (say 100 KB).

Because over 95% of all files are smaller than 100 KB, this would
keep fd usage small, but also keep combined in-transit ( minus
big files) small.

Without big variations of in-transit data, seting for total memory
usage could change. We could specify how much total memory squid
can use. (RAM - few megs). Squid would then be able to dinamicly
set how much memory it can use for "caching" ("caching" = set
memory usage - (program + index + in-transit files)).
If in-transit files sudenly jumped few URLs that were cached
could be dumped. If Memory was scarce, squid would become NOVM.

--MimeMultipartBoundary--
Received on Tue Jul 29 2003 - 13:15:42 MDT