Adrian Chadd wrote:
>
> On Tue, Oct 17, 2000, Henrik Nordstrom wrote:
> > Update is on the way...
>
> Cool. Have you had time to braindump anything else? :-)
No.
> I have
> an idea on an interesting SMP-enabling squid modification which
> doesn't require any shared memory or filedescriptor sharing evilness,
> it depends upon the underlying object storage being able to be accessed
> from multiple processes (so you either have reiserfs_raw or similar
> as the backing store with storetree code in squid, or your squidfsen
> become external processes..)
Sounds like one of the ideas I have..
Run multiple "independent" Squid processes, all using the same backend
store. The only thing the processes share is
a) The backend store
b) The listening socket
c) Some IPC mechanism for sharing the accept() load
Unfortunately raw reiserfs is not entirely suitable yet due to not being
able to differentiate between partially and fully stored objects, but I
imagine it could quite easily be extended to exclusive writer access (no
other readers while writing to an object).
/Henrik
Received on Tue Oct 17 2000 - 07:01:45 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:12:49 MST