Re: [squid-users] [wishlist] removal policy should also move objects to sibling caches

On Fri, 2003-07-04 at 06:25, Bob Arctor wrote:
> it would be great if squids could move data to sibling caches having enough
> free space available instead of merely deleting them.
> this can be done by introducing special icp_balance query .

ICP is one approach. Another would be to provide a generic pre-purge
action, able to use external logic to do whatever is needed.

Extending ICP may not be the best approach:
You'll need some handshaking:
  * you can't purge a candidate until the peers have had time to pick it
up.
  * peers may not *want* it.
 
ICP today is very simple - you will be increasing it's complexity.

That said, I'm not convinced that you need this. caches tend to be more
rather than less full, and content being ejected is that matching your
replacement policy - which is likely to be the same on peers, meaning
that those objects will be immediate candidates for ejection (assuming
you preserve metadata, which would make sense within a cluster).

Rob

-- 
GPG key available at: <http://members.aardvark.net.au/lifeless/keys.txt>.