Ross Esmond Notes

Network Cache

A Network Cache is a local replica of distributed data to facilitate low-latency access. Importantly, a cache is not the source of truth for some data, and could be outdated. If the validity of the data were always checked when the data was accessed, however, that might eliminate any advantage to holding a local copy. A Network Cache must then be designed to accomidate the needs of the particular system in which it resides.

Cached data, which we will call assets, may be accessed by the local program or updated on the source-of-truth. We will refer to the local program that contains the cache as the client and the source-of-truth that provides the assets as the server. Sometimes the client will hold a full replica of all of the servers assets. Conversely, if the cache contains a subset of the available assets, the client may produce cache misses in which it attempts to read an asset that has not yet been cached. The client requires a different set of assets at different times, $N$, and the cache contains a set of assets, $C$. The cache misses that must produce a normal network request are $M = N-C$.