Chapter 8. Intercache Protocols

only for RuBoard - do not distribute or recompile

Chapter 8. Intercache Protocols

In this chapter, I will cover four different intercache protocols. These specialized protocols are used between cooperating proxy caches for a number of reasons. Their most important function is to aid in forwarding decisions. That is, given a request and some number of neighbor caches, to which cache should the request be sent? Or should it be sent directly to the origin server? The intercache protocols provide information that a cache may use to minimize the time to retrieve an object. They may also be used to implement other features, such as prefetching and metadata updates.

The Internet Cache Protocol, or ICP, was the first intercache protocol invented. This lightweight query-response protocol probes neighbor caches for would-be hits. ICP is perhaps the most widely used of the protocols we will discuss in this chapter, mostly because it has been around the longest. All of the other protocols we will discuss were designed to improve upon ICP in one way or another. The most common complaint about ICP is the additional round-trip delay that every request incurs.

The Cache Array Routing Protocol, or CARP, is very different from ICP. It uses a deterministic algorithm, rather than a real-time query, to select a next -hop cache. CARP places emphasis on distributing load among a cache cluster rather than finding a cached copy of a given resource. Unlike some simpler load distribution techniques, CARP nicely accommodates changes in the number of cluster members . In theory, CARP can also be implemented in browsers or other user - agents ”a proxy autoconfiguration script for example.

The Hypertext Caching Protocol, or HTCP, is similar to ICP in that it is also a UDP-based query-response protocol. One major difference is that HTCP uses much more information to determine whether a request would be a hit or a miss . Another significant improvement is the addition of strong authentication. The downside is that the message format is significantly more complex.

Cache Digests are compact representations of a cache's contents. The digest provides a way for a cache to quickly look up an object in its neighbor's directory. Thus, the per-request delays of ICP and HTCP are eliminated. The tradeoff comes in the form of increased memory usage and some amount of inaccuracy in the predictions . Overall bandwidth utilization is probably similar to that of ICP but with different characteristics. Network traffic due to ICP is proportional to the cache request rate and is relatively smooth over time. Cache Digests, on the other hand, generate significant bursts in network utilization at regular intervals.

If you are looking for WCCP here, you won't find it. WCCP is not used between cooperating caches. Rather, it is a protocol that switches and routers use to divert traffic to a proxy cache. See Section 5.2.3, for the scoop.

only for RuBoard - do not distribute or recompile