Re: squid-1.1.22 performance

Dave Zarzycki writes:

>Duane Wessels,
>
>We (Rhapsody @ Apple) have been experimenting with squid-1.1.22 as a HTTP
>accelerator with excellent results. But, we *think* it should be able to
>go faster. From a debug dump, we see lots of this:
>
>1998/08/19 10:18:07| memCopy: offset 8192: size 4096
>1998/08/19 10:18:07| comm_write: FD 13: sz 4096: tout 30: hndl 0x1b53c:
>data 0x4fa650.
>
>
>Two questions come to mind:
>
>1) Why the memCopy if it's already in memory? (i.e. small files that
>should fit into the cache mem area)

part of it is just historical/evolutionary. it just works that way.

once answer might be that it saves us from having to lock areas of
memory. The buffer might not be written right away, so it would be
bad if the store.c functions freed the memory before we wrote it.

>2) Why only write 4k to the socket?

I guess it just feels like a reasonable number. Plus, in our
experience, half of the cache objects are less than 4k.

Duane W.
Received on Wed Aug 19 1998 - 13:49:59 MDT