47 results about "I/O scheduling" patented technology

The invention provides a RAID-6 I/O scheduling method for balancing strip writing. The method comprises: reorganizing write requests by analyzing input/output distribution of the requests; minimizing the number of modifications of corresponding check by means of check sharing; merging input/output fragments and a write request in an incoming same strip into a new write request; writing the new write request to a disk once; marking an illegal data block generated in a reorganization process; recording the illegal data block in a memory; performing trash recycling on the illegal data block; and storing metadata information of the illegal data block for redistribution preparation. A write operation can be reduced, and disk load in a RAID storage system is balanced. Basic ideas of the invention is to perform strip writing on the basis that the corresponding check is globally understood, and to reduce the number of check modifications required for strip writing by reorganizing distribution of data blocks.

The invention discloses a method for separately processing data reading and writing requests and metadata reading and writing requests. The method comprises the following steps: adding metadata semantics of requests; setting a data request queue and a metadata request queue for each SCSI (small computer system interface) equipment, wherein the metadata request queue is especially used for processing metadata requests, and different processing modes are adopted for different requests. The metadata request queue adopts a priority response processing mode according to the characteristics that the metadata requests are small in particle size, high in instantaneity and disperse in distribution; the requests do not pass through an I/O (input/output) scheduling layer, are not delayed in the request queue, and are directly issued to a disk driver; the data requests which are large in request particle size, relatively low in frequentness, and centralized in distribution adopt the mode that the requests pass through the I/O scheduling layer, are stored in the request queue for a period of time, and are issued to the disk driver after being discharged, so that the bottleneck problem of a single request queue is overcome; the parallelism of different levels of request processing is strengthened; the method is especially suitable for the business which needs to frequently and greatly store small files, and has high demands on critical data instantaneity.