next up previous
Next: CRISP structures Up: Directory Structures for Collective Previous: Directory Structures for Collective

Hierarchical ICP Caches

Squid and other Harvest-derived hierarchical caches maintain no directory apart from each caching server's directory of its own contents; the global directory consists of the aggregate of the local directories. Caches are organized in a loose hierarchy: to probe the collective cache, a caching server multicasts an ICP message to its neighbors (siblings) and parents, which respond with a HIT only if they have a local copy of the requested object. If the request misses or times out, a selected parent continues the probe by repeating the process, multicasting to its neighbors and parents. If the request misses at each level of the hierarchy, then a top-level parent fetches the object from the source, caches it, and passes it back down the hierarchy.

This approach has several drawbacks:

These problems motivate ``flat'' cache structures with a scalable directory structure that provides a more complete index of the collective cache's contents.


next up previous
Next: CRISP structures Up: Directory Structures for Collective Previous: Directory Structures for Collective
Syam Gadde
1998-05-19