Robert Collins wrote:
> I envisage the follow modules to start with:
> storefs - coss, ufs, null, sfs, reiserraw
> storeio - ufsio, aufsio, diskd
Fine, but I am a little sceptic about the ufs* names.. and even more
sceptical that such an interface can be done properly in a
time-efficient manner at present.
For example: The current interface between aufs -> async-io (which is an
API of it's own) is quite bad and is long overdue a complete redesign.
The diskd interface shares some of the same problems in the fs->io
interface.
See my old notes on http://squid.sourceforge.net/hno/async-io.html
This is also partly true for the Squid->FS interface, which is slowly
getting in shape.
I see it as utterly more interesting to get the FS interface in a good
shape (what Adrian is attempting in "modio") to be able to support the
varity of filesystems there might be, and to try to optimize the FS
types we have and/or add a couple of new ones for better performance.
When we have a couple of efficient stores, starting to separate the
FS:es into "logics" and "io" might be interesting.
Third, the rebild logic in "ufs" is not a great one to reuse for
building new FS:es...
/Henrik
Received on Sun Feb 18 2001 - 01:48:26 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:13:32 MST