The Design of a Simulation System for Persistent Object Storage Management

EEcient storage management of objects is an important part of any persistent object system. Storage management algorithms, such as those for clustering, caching, and garbage collection, are often complex, since they must simultaneously satisfy a number of constraints, including low CPU overhead, minimum space usage, and low latency. We have designed a simulation system that can be used to evaluate a wide variety of such algorithms. The system is innovative in that it uses trace-driven simulation as well as a loosely-coupled architecture that separates out diierent aspects of storage management policy, machine hardware, and simulation control into separate modules. Examples of such modules are those for storage allocation, object connectivity, physical characteristics, and cost modeling. The modules communicate using message multicast (i.e., selective broadcast). This approach supports rapid prototyping and rapid evaluation of alternative storage management algorithms.