Electrical computers and digital processing systems: processing – Dynamic instruction dependency checking – monitoring or... – Reducing an impact of a stall or pipeline bubble
Reexamination Certificate
1998-03-13
2001-09-18
Hoff, Marc S. (Department: 2857)
Electrical computers and digital processing systems: processing
Dynamic instruction dependency checking, monitoring or...
Reducing an impact of a stall or pipeline bubble
C709S241000
Reexamination Certificate
active
06292885
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a method of estimating the effect of parallel processing achieved by a computer and a recording medium used for the method. In particular, the present invention relates to a method of estimating parallel-processing time required by a computer to process steps of a procedure in parallel with the use of a pipe data set. The pipe data set is prepared in a system memory and is used to transfer data among the steps of the procedure to be processed in parallel. Based on the estimated parallel-processing time, the method estimates the effect of the parallel processing of the steps. The present invention also relates to a recording medium used for the method.
2. Description of the Related Art
A JCL (job control language) describes a job to be processed by computer. Each job consists of procedures that are called job steps. Each procedure, i.e., job step consists of steps. Steps and job steps are processed sequentially in described order. Any computer capable of processing instructions in parallel processes job steps in parallel to improve efficacy. If a data set provided by a first job step is required by a second job step that follows the first job step, the data set is stored as a pipe data set in a virtual file in system memory instead of being stored in a direct access storage device or a magnetic tape device. A given job is usually processed through a batch process, and a batch process that uses such a pipe data set is called an excel batch in this specification, and a batch process that does not use it is called a normal batch.
Japanese Unexamined Patent Publication No. 9-6628 of this applicant discloses a method of and an apparatus for selecting procedures to be processed in parallel. This disclosure examines at least one data set in consecutive procedures to see how the data set is used and selects procedures that are processible in parallel as a data set used for a pipe data set. The pipe data set will be explained.
FIG. 12
shows a pipe data set used to transfer data among job steps that are processed in parallel. The pipe data set is a system memory file into which a job step
1
writes data and from which job steps
2
and
3
read data.
The job step
1
provides data pieces B
1
to Bn, which are immediately transferred to the job steps
2
and
3
through the pipe data set. The pipe data set is then cleared, so that a small area of 64 kilobytes to several hundreds of kilobytes for the pipe data set in a system memory is effectively used to transfer even a large amount of data pieces.
The space for the pipe data set is secured in the system memory whenever one of the job steps
1
to
3
first opens the pipe data set. When none of the job steps open the pipe data set, environments for accessing the pipe data set are prepared.
If, for example, the job step
2
tries to read data and if the job step
1
has not yet written data into the pipe data set, the job step
2
must wait for the job step
1
to write the data. When the job step
1
writes data pieces B
1
to Bn into the pipe data set, the pipe data set is saved as a temporary system memory file.
The job steps
2
and
3
immediately read the data pieces in order of B
1
to Bn. Namely, the job steps on the reading side receive identical data. As soon as the data pieces are completely read, the pipe data set is cleared to free the space. This space is cyclically used.
If the pipe data set is full, the job step
1
must wait to write data until the job steps
2
and
3
read data to free the space of the pipe data set. If the pipe data set is empty, the job steps
2
and
3
must wait to read data until the job step
1
writes data into the pipe data set. In this way, data transfer between the job step
1
and the job steps
2
and
3
is carried out in parallel asynchronously.
If some job steps are processible in parallel in an excel batch with the use of a pipe data set, it is necessary to estimate an efficiency improvement realized by the parallel processing of the job steps compared with the serial processing thereof. Namely, it is necessary to estimate parallel-processing time for processing the job steps in parallel and time reduction produced by the parallel processing of the job steps. The prior art forces the user to manually calculate the parallel-processing time and time reduction according to SMF (System Management Function) records of the job steps. The SMF prepares a job schedule and collects, as SMF records, various event data such as job start time and job end time. The prior art needs a long time and much labor to manually calculate the parallel-processing time and time reduction.
SUMMARY OF THE INVENTION
An object of the present invention is to provide a method of automatically estimating the effect of processing steps of procedures in parallel by a computer and a recording medium used for the method.
FIG. 1
shows a basic arrangement of the present invention to accomplish the object. An apparatus
1
for estimating the effect of processing steps of procedures in parallel is realized with a computer. An SMF data set
2
is in a storage device such as a magnetic disk and contains SMF records prepared when processing the steps in series. The apparatus
1
collects the SMF records, analyzes them, calculates series-processing time for processing the steps in series, estimates the time reduction produced by processing the steps in parallel, and estimates parallel-processing time by subtracting the time reduction from the series-processing time, thereby estimating the effect of the parallel processing of the steps.
The method may edit the parallel-processing time and time reduction and output an estimated effect list
5
through a printer or a CRT.
In this way, the method automatically estimates and outputs the effect of processing steps of procedures in parallel by computer.
The present invention also provides a method of estimating the effect of processing first and second procedure groups in parallel by a computer. The first procedure group transfers data through a pipe data set, and the second procedure group transfers no data through a pipe data set. The method includes the steps of collecting SMF records prepared when processing the first and second procedure groups in series, calculating, from the SMF records, series-processing time for processing the first and second procedure groups in series, summing up estimated processing time of each procedure of the first procedure group according to the SMF records, to provide a first processing time for processing the first procedure group in parallel, choosing a maximum among the values of estimated processing time of procedures of the second procedure group according to the SMF records, to provide a second processing time for processing the second procedure group in parallel, choosing a maximum between the values of the first and second processing times, to provide parallel-processing time for processing the first and second procedure groups in parallel, and estimating time reduction produced by the parallel processing of the first and second procedure groups by subtracting the parallel-processing time from the series-processing time, thereby estimating the effect of processing the first and second procedure groups in parallel.
The method may further have the steps of editing the parallel-processing time and time reduction and outputting an estimated effect of parallel processing through a printer or a CRT.
The procedure groups may form a composite procedure. Series-processing time of the composite procedure is calculated by summing up the parallel-processing time of each of the procedure groups.
A procedure may consist of first and second groups of steps. The first group of steps transfer data through a pipe data set, and the second group of steps transfer no data through a pipe data set. In this case, the present invention provides a method of estimating the effect of processing the first group of steps in parallel by computer. The method includes the steps of collectin
Honda Tamotsu
Komura Makoto
Nakai Hajime
Fujitsu Limited
Hoff Marc S.
Miller Craig Steven
Staas & Halsey , LLP
LandOfFree
Method of estimating effect of parallel processing and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method of estimating effect of parallel processing and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of estimating effect of parallel processing and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2504404