>
> > COSS is a neat idea (thaanks Eric!) and it doesn't have the problem of
> > a straight cyclic store.
>
> Eric stole this idea from me ;-) (see the early discussions on
> squid-dev, partly archived on my web pages). Thanks Eric for
> implementing it.
Yup, I remember your ideas from reading the squid-dev archives
on your webpage.
> > It is full of memcpys, limited to 2gb stores
> > currently, and is not asyncio ready, but as it was a nice tease for
> > the storage io /replacement reorganisation, it will be a nice tease
> > for the storage manager memory reorganisation (one of the thing I
> > want to do with COSS and UFS is support clustered writes so COSS
> > doesn't require a membuf and lots of copying and UFS objects are written
> > in chunks larger than DISK_PAGE_SIZE or SM_PAGE_SIZE).
>
> Ok. If you are working on the store memory management, then I'll take a
> look into cleaning up the replacement policy management.
Please. I'll probaby take a while doing this, but I believe it will
be worth it in the end for more reasons than async disk IO.
Adrian
Received on Sun Apr 23 2000 - 10:54:38 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:12:23 MST