Squid memory footprint (was: MemPools rewrite)

 I wonder how can you eliminate StoreEntry? IMHO it contains crucial
 information that allows squid to skip disk accesses. Moving parts
 of this data into squidfs doesn't seem to change much in ram usage.
 Moving this crucial information onto disks implies enormous performance
 penalty, doesn't it?

On 1 Nov 2000, at 23:36, Henrik Nordstrom <hno@hem.passagen.se> wrote:
> Alex Rousskov wrote:
> > On Wed, 1 Nov 2000, Henrik Nordstrom wrote:
> >
> > > And there currently is at least one implementation using "the right
> > > approach", based on a variant of the reiserfs filesystem on Linux. In
> > > there the whole cache maintenance is moved to the file system. Squid
> > > only stuffs content there and retreives objects as needed.
> >
> > Is memory-resident StoreEntry for each object eliminated with
> > reiserfs? If yes, that's very cool, and I am obviously way behind on
> > the developments.
>
> Yes. But as of yet it is done in a quite crude way. Adrian has volunteer
> to look into the store interface if a good design can be found for
> supporting stores where a in-core index is not needed.
>
> However, for the reiserfs_raw (aka butterfly) store, as there is no
> in-core index at all there is a farily big penalty for misses in
> querying the disk storage to see if it could be a hit..
>
> /Henrik

------------------------------------
 Andres Kroonmaa <andre@online.ee>
 Delfi Online
 Tel: 6501 731, Fax: 6501 708
 Pärnu mnt. 158, Tallinn,
 11317 Estonia
Received on Thu Nov 02 2000 - 01:00:25 MST