Dismiss
InnovationQ will be updated on Sunday, Oct. 22, from 10am ET - noon. You may experience brief service interruptions during that time.
Browse Prior Art Database

Threshold Scheduling Policy for Mirrored Disks

IP.com Disclosure Number: IPCOM000102621D
Original Publication Date: 1990-Dec-01
Included in the Prior Art Database: 2005-Mar-17
Document File: 2 page(s) / 92K

Publishing Venue

IBM

Related People

Thomasian, A: AUTHOR

Abstract

Disclosed is a scheme to improve the efficiency of write operations in the case of replicated data on mirrored disks, which is based on a threshold scheduling policy with respect to the number of accumulated write operations. This scheme should result in a significant increase in the bandwidth of a mirrored disk configuration.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Threshold Scheduling Policy for Mirrored Disks

       Disclosed is a scheme to improve the efficiency of write
operations in the case of replicated data on mirrored disks, which is
based on a threshold scheduling policy with respect to the number of
accumulated write operations.  This scheme should result in a
significant increase in the bandwidth of a mirrored disk
configuration.

      Mirrored disks are a well known method to achieve high data
availability in coping with disk media failures.  They have been
implemented in several computer systems, including those of Tandem
(1) and IBM (ESA/3090* series (2)).  Note that the disks in the two
cases are not synchronized (3) but the layout of files on both disks
is expected to be the same.

      Disk mirroring provides a significant increase in the disk IO
bandwidth available for read operations, since essentially we have a
queueing system with two servers.  At lower disk utilizations
improvements in disk service time (by reducing the seek time) are
reported in the context Tandem machines by dedicating the read/write
mechanism on each disk to disjoint and contiguous subsets of the
cylinders on the disks.

      In recent years there has been an increased demand for the
processing of read-only queries in a DBMS which also processes update
transactions.  It is shown in (4) that the disk IO bandwidth
available for query processing decreases rapidly with the increase in
update transaction disk accesses.  In this disclosure we concentrate
on the interference between the two workloads at the disk IO level.

      This disclosure is based on a particular scheduling policy,
which was described in the context of the reader's and writer's
problem in a multiprocessor environment [5]. It was shown that
adopting this policy results in a significant increase in throughput
compared to a FCFS (first come, first served) policy, especially when
read operations are much longer than writers.

      The problem solved by the invention can be described as
follows.  Apply the threshold scheduling policy to read and write
operations on mirrored disks by deferring disk writes, while there
are read operations in progress until a certain threshold (in the
form of number of writes pending or buffer space utilizatio...