Importance

Loading Importance Indicators...

Abstract

A system and method for reducing latency in systems using layered device driver architectures. The preferred embodiment of the invention comprises a disk subsystem including a layered device driver architecture which performs various RAID operations. A plurality of the driver layers are defined to use a common request structure. The system and method of the present invention determines if a request maps unmodified to a lower driver layer and 'quick launches' these requests to the lower driver layer according to the present invention. When a host request is received, the first layer device driver executing on the disk controller determines if the enqueued request is atomic, i.e., if the request maps unmodified to the next lower level driver layer. If so, then the request is dequeued from the first level queue and enqueued on the next lower level layer queue, avoiding steps of allocating and populating a new request structure for the lower level driver. This process preferably repeats for each of the device driver layers. When the last device driver layer completes, each device driver performs a completion callback to its calling driver layer. If a respective device driver layer was quick launched from a higher level driver layer, bypassing one or more intermediate driver layers, then the respective device driver layer performs a completion callback to the higher level driver layer, also bypassing the intermediate layers. Therefore, the system and method of the present invention reduces and/or virtually eliminates added latency in layered device driver architectures for those requests which pass through one or more device driver layers unmodified.

Loading the Abstract Image...

First Claim