We describe a real implementation of a software component that manages caching of files from a tertiary storage management system to a large disk cache developed for use in the area of High Energy Physics (HEP) analysis. This component, called the Cache Manager, is a part of a Storage Access Coordination System (STACS), and is responsible for the interaction with a mass storage system that manages the robotic tapes (we used HPSS). The Cache Manager performs several functions, including managing the queue of file transfer requests, reordering requests to minimize tape mounts, monitoring the progress of file transfers, handling transient failures of the mass storage system and the network, measuring end-to-end file transfer performance, and providing time estimates for multi-file requests. These functions are described in detail, and illustrated with performance graphs of real-time runs of the system.
Luis M. Bernardo, Arie Shoshani, Alex Sim, Henrik