Browse Prior Art Database

Providing Extra-Surfaces to Decrease Write Commit Times in Disk Subsystems

IP.com Disclosure Number: IPCOM000038896D
Original Publication Date: 1987-Mar-01
Included in the Prior Art Database: 2005-Feb-01
Document File: 3 page(s) / 53K

Publishing Venue

IBM

Related People

Franaszek, PA: AUTHOR [+2]

Abstract

This invention decreases commit times of write requests in disk subsystems by providing extra surfaces reserved for write requests. The response time of an I/O request in the disk subsystem consists of several components: (1) waiting time until the device becomes free, (2) waiting time until a path to the device becomes available, (3) seek time of the arm to the specified cylinder, (4) latency time for the angular position of the device to reach the specified sector, (5) Rotational Position Sensing (RPS) delay in order to reconnect to the channel, and (6) the actual data transfer time. There are two types of I/O requests: a read request and a write request. In a non-cached (native) control unit environment, both read and write requests are served in the same manner except for the direction of data movement.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 3

Providing Extra-Surfaces to Decrease Write Commit Times in Disk Subsystems

This invention decreases commit times of write requests in disk subsystems by providing extra surfaces reserved for write requests. The response time of an I/O request in the disk subsystem consists of several components: (1) waiting time until the device becomes free, (2) waiting time until a path to the device becomes available, (3) seek time of the arm to the specified cylinder, (4) latency time for the angular position of the device to reach the specified sector, (5) Rotational Position Sensing (RPS) delay in order to reconnect to the channel, and (6) the actual data transfer time. There are two types of I/O requests: a read request and a write request. In a non-cached (native) control unit environment, both read and write requests are served in the same manner except for the direction of data movement. In a (volatile) cached control unit environment, on the other hand, a write request is committed only after data have been written to (non-volatile) disk storage. Writing to (volatile) cache may be insufficient to commit the write operation since the

(Image Omitted)

cache is subject to failure and therefore its contents may be lost. We are interested in decreasing the commit time of write requests, i.e., the write response time. Without loss of generality, it may be assumed that an entire track of data is the unit of data transfer between control units and devices. A primary feature of this invention is to provide a non-volatile area in the disk subsystem with shorter response time than that of existing systems. This allows committing a write request after the track has been written to this additional non-volatile storage area. The actual writing of data to its final location on disk occurs as a background, low priority operation. The additional non-volatile storage could be an extra-surface per disk or a reserved surface in existing systems. In this proposed disk configuration, a write request forks into two requests: (1) a write request to an extra-surface which we call fast write, and (2) a conventional write request which we call normal write . A fast write request does not specify a particular device and therefore could be served by any device that is available and preferably one which has its arm located on a free track on the extra-surface of the device. As a consequence, the response time of a fast write is reduced to one delay component, namely, the actual data transfer time. A normal write, on the other hand, is carried out in parallel with fast writes, but at a lower priority, and it encounters the six delay components mentioned...