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

Technique for Improving Performance of Global Resource Serialization

IP.com Disclosure Number: IPCOM000103845D
Original Publication Date: 1993-Feb-01
Included in the Prior Art Database: 2005-Mar-18
Document File: 2 page(s) / 96K

Publishing Venue

IBM

Related People

Blank, TE: AUTHOR [+4]

Abstract

A method is disclosed which improves the responsiveness of global resource serialization requests in an MVS/ESA* system while reducing the amount of processing overhead required.

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

Technique for Improving Performance of Global Resource Serialization

      A method is disclosed which improves the responsiveness of
global resource serialization requests in an MVS/ESA* system while
reducing the amount of processing overhead required.

      Global Resource Serialization (GRS) is a component of the
MVS/ESA operating system which provides programs with the ability to
serialize the use of resources.  An example of such a resource is a
file on direct access storage.  MVS/ESA services exist to perform the
serialization either within a single system or (where multiple
systems have access to the resource) across several systems.

      GRS performs multiple-system serialization by connecting
systems in a ring, and passing a "token" (called the Ring System
Authority message, or RSA) around the ring.  Requests for global
(multi-system) serialization, called ENQs, must wait for the RSA to
arrive at the requesting system, then wait for the RSA to
circumnavigate the ring to insure that no other system has already
received a request for or granted access to the resource being
requested.

      Processor cycles are required on each system to receive,
process and send the RSA around the ring.  A user definable amount of
time that the RSA should spend waiting in each system has always been
part of the GRS external specifications.  This delay is termed
RESMIL.  By choosing the right value for RESMIL, customers could
control the trade-off between better response time for global
serialization requests (lower RESMIL) and smaller CPU overhead for
GRS ring processing (higher RESMIL).  However once set, this value
was difficult to change and unresponsive to differing system
workloads where one setting might be preferred over another.  RESMIL
is currently specified in milliseconds.

      This disclosure describes a technique for dynamically
regulating the performance of a multi-system GRS ring to achieve
acceptable ENQ response time while minimizing GRS CPU overhead at
times when ENQ traffic is lighter and response time is less critical.
This technique is termed Self-Tuning RESMIL.

      Self-tuning RESMIL considers the value of RESMIL specified by
the customer as a minimum rather than a fixed value.  An upper bound
(and thus the range over which the system can vary RESMIL) is defined
by the system, and may vary from release to release based on analysis
of workloads.  This technique can also be used in other scenarios and
applications where an interval between time-driven events is adjusted
based on past-performance, present workload and predictions of future
requests.

The self-tuning RESMIL algorithm is defined as follows.

o   Each system monitors the RSA for any...