Disk reads with DRAM latency
Garth A. Gibson, Russel H. Patterson, Mahadev Satyanarayanan · 2003
To overcome the relative deterioration of secondary storage performance, input/output systems are being designed around highly parallel disk arrays, high-bandwidth optical networks, and cache-specialized file systems. Disk arrays increase secondary storage throughput by spreading data over many disks so that large accesses are striped for parallel operation and so that many small accesses can operate concurrently. Fast networks preserve throughput improvements derived from disk arrays on file server as data is delivered to increasingly distant client systems. Secondary storage latency is primarily addressed by large client and server file caches. Whenever data to be read is found in the cache or data to be written can be delayed in the cache, latency is as short as if secondary storage were constructed with semiconductor memory. The current research focuses on reducing the access latency for data to be read that is not already cached.>