Stripe Cache I/O Management for RAID Systems

The patent applies preliminary action by pre-reading data from disk into cache memory before actual I/O requests need to access it. When the system detects that multiple I/O requests may target the same stripe, it proactively loads the stripe data into cache, so that subsequent requests can be served from cache rather than causing disk access conflicts. This anticipatory caching resolves the contradiction by preparing resources in advance to enable parallel processing without actual disk conflicts.

The cache memory serves as an intermediary between the parallel I/O processing system and the physical disk. Instead of allowing multiple parallel requests to directly conflict on disk access, the cache acts as a buffer layer that can satisfy multiple requests simultaneously from its stored data, thereby mediating the conflict and enabling parallel processing to proceed without actual disk access interference.

The patent merges multiple I/O requests that target the same stripe into a single cache lookup operation. Instead of allowing each request to independently access the disk (causing repeated reads), the system combines these requests and serves them all from the cached copy of the stripe data. This merging approach maintains data redundancy requirements while eliminating the performance penalty of repeated disk accesses.

The patent creates a copy of the stripe data in cache memory that can be served to multiple I/O requests simultaneously. This copying mechanism allows the system to maintain the necessary data redundancy for reliability while providing fast duplicate copies to multiple consumers without requiring repeated accesses to the original disk, thereby resolving the contradiction between reliability and productivity.