Henrik Nordstrom wrote:
>
> Ernst Heiri wrote:
>
> > What's the difference between 1.1.X and 1.2 causing this ?
> > (I'm still using 1.1.20 and have no experience with 1.2)
>
> The major difference is that 1.2 maintains each cache directory
> separately. You can add/remove cache directories as needed without
> wiping your cache. Squid automatically adjusts and balances the usage on
> the available cache directories.
Erm, teensy terminological note...I think it would be less confusing to
say 'cache-tree' or (better, IMO) use the keyword 'cache_dir', since
some people probably confuse the cache-buckets with the term 'cache
directory'.
> Performance differences:
> * There is is ongoing work to balance the I/O load evenly on each cache
> directory.
Mmm. And it's GOOD(!) :)
> * Squids I/O pattern is more suited for one cache directory per disk
> than a stripe. Squid primarily needs to be able to read/write at many
> different places at the same time (==lots of seeks) and keeping one
> operation/object isolated to one disk gives fewer seeks in total. Squid
> does NOT need a very high I/O rate on one single request.
Indeed. Why hash in userspace _and_ then stripe in kernelspace? Seems
like a big waste of cycles, that can be better spent serving users. No
need for raid and no funny business adding cache_dir's. Each can also be
sized individually with 1.2 (join me in a sigh of relief).
D
-- -----BEGIN GEEK CODE BLOCK----- Version: 3.1 GAT d- s++: a C++++$ UL++++B+++S+++C++H++U++V+++$ P+++$ L+++ E- W+++(--)$ N++ w++$>--- t+ 5++ X+() R+ tv b++++ DI+++ e- h-@ ------END GEEK CODE BLOCK------Received on Sat May 16 1998 - 06:58:17 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:40:12 MST