IBM System/370 Guide

JOB SCHEDULING
The following are guidelines for job scheduling.
1. If total system throughput improvement, rather than maximum
individual job performance increase, is the objective, schedule
jobs that use RPS together. such th~t the maximum number of
RPS devices are active concurrently on each block multiplexer
channel.
The more RPS devices active concurrently on a block multiplexer
channel, the greater the overlap potential that exists. <See
item 1 under -System Configuration and Generation-.)
2. When jobs that use disk data management functions with RPS
support are executed concurrently with jobs that do not use
RPS support, the devices assigned to the former should be on
different channels from devices assigned to the latter.
If jobs using RPS and jobs not using RPS must access devices
on the same block multiplexer channel concurrently. the jobs
without RPS support should have high seek activity such that
search and read/write time is small compared to seek time.
Assume sequentially organized data sets and TCAM message queues
are allocated to the same 3330 facility on a block multiplexer
channel. A QSAM job step and a TCAM job step are executing
and access the 3330 facility concurrently. Since RPS commands
are not used for TeAM message queue processing, each 3330 disk
drive containing TeAM queues acts like a non-RPS device and
can monopolize channel time. Thus, the job steps that use QSAM
can be delayed. <See item 2 under -System Configuration and
Generation-.)
3. Allocate a data set that will be accessed using QSAM or BSAM
chained scheduling to a device on a channel without active RPS
jobs.
The chained scheduling technique is designed to keep a device
active as long as record processing keeps up with record reading
or writing. Thus. either a selector or a- block multiplexer
channel can be kept busy for long durations, preventing the
execution of any other I/O operation on the channel. Note that
while QSAM and BSAM support concurrent use of RPS and chained
scheduling for access to a disk data set, the performance
attained by using chained scheduling alone will not be improved
significantly by using RPS as well.
4. When data sets are being processed by an RPS access method in
a multiprogramming or multitasking environment and disk device
assignment is handled by the user rather than by the control
program, allocate as many separate RPS devices as is practical.
This approach allows the possibility of having more concurrent
operations on these data sets and therefore more seek and
rotational positioning overlap.
5.. If a response-oriented RPS job operates on a block multiplexer
channel concurrently with other RPS jobs, job scheduling should
insure that the number of jobs executing simultaneously is such
that the performance desired for the response-oriented job can
be attained.
The performance of a block multiplexer channel is affected by
the percentage of time the channel is busy searching and reading.
91

IBM System/370 Specifications

General

Brand	IBM
Model	System/370
Category	Server
Language	English