Text preview for : B323 JES2 JCL Error Study and Requirements; Mosteller.pdf part of IBM B323 JES2 JCL Error Study and Requirements; Mosteller IBM share SHARE_61_Proceedings_Volume_1_Summer_1983 B323 JES2 JCL Error Study and Requirements; Mosteller.pdf
Back to : B323 JES2 JCL Error Study | Home
In our installation, our budget is not always sufficient
for our requirements. That could be true in most installations,
but in smaller shops the percentage of the data processing budget
SESSION REPORT
lil
SHARE NO.
B323
SESSION NO.
J E S 2
SESSION TITLE
Wi 11 i amMo s t e 11 e r
--=
~SHARE~
JES2: JCL Error Study and Requirements Potpourr_i__~6=5_______
ATTENDANCE
BC.G
required for software would always be higher than a larger installa- PROJECT SESSION CHAIRMAN INST. CODe
tion, since basically all the same software is required. Because
of this, either we do not always acquire the software we should Boei~g Computer Services Company. 7990 Gallows Court, Vienna, VA 22180, 703 821 6208
have, or we support 'home-grown' or 'free' software which requires seSSION CHAIRMAN'S COMPANY. ADDRESS. and PHONE NUMBER
more maintenance. As an example, IBM over the last few years has
added different program products to the full MVS IPO system. In
themselves, some of these products are not that expensive and most
large installations already have them. But in our case, each time
we have to justify this new product. Currently, we use a segmented
IPO, not the full IPO, because of this. As well, monies to send
technical staff on courses and to make a regular commitment to
conferences we should be attending is a problem. JES2: JCL Error Study and Requirements Potpourri
Each systems programmer in a small MVS shop usually is William Mosteller (BCG)
responsible for the support of many different systems or products.
There is not a situation where one or two people or even a group Session B323, SHARE 61
can specialize in a specific area. In this way, there is very
little backup for a person, if any. It is very difficult to free Wednesday, August 24, 1983, 1:00 PM
staff for new projects because they are too busy, and if a pro-
grammer leaves, the shop usually is in difficulty for a given
period of time. The software environment can also be just as com-
plex as in larger MVS shops. These complex smaller shops may have
~
more manpower than some other smaller shops, but here each individual
00 is heavily relied on to support multiple systems as well.
N
Tuning an MVS system for a small user is different than
for a large user. You have similar tuning concerns, but in a small
shop you do not normally have all the tuning aids and monitors.
Also, you do not have the luxury to dedicate volumes for swapping
and paging. Usually you are restricted by a limited amount of DASD
to perform the proper reconfigurations that you would like to make.
These are some of the concerns I have come across as
a small MVS user.
CONCLUSION
In conclusion, we feel that MVS can run quite well on
a 3031 CPU, but like any other CPU that MVS runs on, MVS' per-
formance is very dependent on the amount of real memory the com-
puter has and the amount of DASD available to properly layout
paging and swapping data sets, and application volumes. Without
sufficient memory and with a limited amount of DASD, we found
that we have to restrict the use of the computer resources,
especially during the prime shift when you are trying to provide
an acceptable on-line response for your terminal users. We have
had to cut hack on our batch users to provide this response. This
implies execution of their jobs during off-shifts and a reduction
in the number of jobs submitted.
~l-TP".'" _ i ~ n _~ l Q ,
JCL Error Study - 2 - SHARE 61 JCL Error Study - 3 - SHARE 61
At the session today I hope to accomplish four things: 1) present -,
the effect of a JCL error improvement in MVS SP 1.3.3, 2) discuss a Symbol Not Defined in Procedure (IEF657) I
JES2 change which supports automatic spelling correction, 3) show you Error Frequency Improvement I
the results of the JCL error survey, and, finally, 4) discuss JES2 I I
requirements for which no other working session is available. The Message Before After I Change I
shortage of meeting rooms at this SHARE left us no choice but to April, 1982 June, 1983 (Week) I I
include that last item with the first three. I I
IEF601 31 0.9% 2 0.5% -0.4% I
We are very lucky to have two I BM representatives from the Con- IEF603 1 0.0% 0 0.0% -0.0% I
verter/Interpreter here today. Leslie Dumas (IBM), as many of you IEF605 945 28.4% 129 31. 7% 3.3% I
will remember, discussed the Scheduler JCL Facility 6 months ago in IEF607 178 5.4% 24 5.9% 0.5% I
San Francisco. Sandy Canetti (I BM) is, I believe, a new attendee at IEF610 6 0.2% 0 0.0% -0.2% I
SHARE. [Give them the floor briefly.] IEF611 78 2.3% 11 2.7% 0.4% I
IEF612 74 2.2% 6 1.5% -0.8% I
JCL Error Improvements in MVS/SP 1.3.3 IEF613 8 0.2% 0 0.0% -0.2% I
IEF617 3 0.1% 0 0.0% -0.1% I
As most of you know, I have been collecting JCL error messages for IEF618 83 2.5% 8 2.0% -0.5% I
about two years, both at my own site and at a couple others. Across IEF621 478 14.4% 59 14.5% 0.1% I
the sites, we observed that 9 error messages represent around 85% of IEF622 94 2.8% 6 1.5% -1.4% I
the messages JCL coders see, and we developed a requirement for IBM IEF623 103 3.1% 17 4.2% 1.1% I
improvements to those messages. IEF624 144 4.3% 29 7.1% 2.8% I
'IEF625 52 1.6% 4 1.0% -0.6% I
One of the messages, "Symbol Not Defined in Procedure" (IEF657), IEF626 1 0.0% 0 0.0% -0.0% I
was dramatically improved in MVS/SP 1.3.3. Before, the message did IEF627 42 1.3% 13 3.2% 1.9% I
...
not report the symbol the converter found offensive and did not tell IEF628 9 0.3% 6 1.5% 1.2% I
where it found the symbol. Now, it provides both these vital facts IEF629 44 1.3% 8 2.0% 0.6% I
about the error. Because improved specificity makes correction easier, IEF630 243 7.3% 32 7.9% 0.6% I
00
I expected that the improved error message would reduced occurrences IEF632 227 6.8% 33 8.1% 1.3% I
'" of the message. I expected a reduction because difficult messages
cause users to resubmit their jobs without change, or with ineffective
IEF641
IEF642
21
14
0.6%
0.4%
4
1
1.0%
0.2%
0.4%
-0.2%
I
I
changes. Improved specificity, I reasoned, would increase the chances IEF647 3 0.1% 0 0.0% -0.1%
of a successful correction the first time. IEF650 2 0.1% 0 0.0% -0.1%
IEF651 16 0.5% 1 0.2% -0.2%
John Kinn (CL), of Lockheed California Company, has MVS/SP 1.3.3 IEF652 83 2.5% 4 1.0% -1.5%
and my JCL error collection modification installed on his system. He IEF657 326 9.8% 7 1.7% -8.1%
was kind enough to provide me with data. Before, in March, 1982, IEF658 5 0.2% 2 0.5% 0.3%
9.8% of his messages were "Symbol Not Defined in Procedure." During IEF663 2 0.1% 0 0.0% -0.1%
a week in June, 1983, that number dropped to 1.7%, representing a IEF668 11 0.3% 1 0.2% -0.1%
shift of 8.1%. This shift is enormous. Without MVS SP 1.3.3, "Symbol
Not Defined in Procedure" represents 8.6% of the error messages at TOTAL 3327 100% 407 100%
BCG, and 9.4% of the error messages at CSG (Canada Systems Group,
data courtesy of Terry Burr (CSG)). Thus, the greatest difference
among the three sites was 1.2%. Compared with these numbers, the
8.1% shift at Lockheed remains enormous.
JES2 Automatic Spelling Correction
I conclude that improving error messages makes JCL easier to use,
and recommend that you install MVS/SP 1.3.3 simply for the benefits of Howard Morgan published a spelling correction technique in the
the improved message. "Communications of the ACM" in February, 1970. The technique cor-
rects keywords with anyone of four possible errors:
August 15, 1983 August 15, 1983
JCL Error Study - 4 - SHARE 61 JCL Error Study - 5 - SHARE 61
Single missing letter, JCL Error Survey Results
Single added letter, In hopes of improving the business case for IBM improvements to the
JCL converter, we conducted a su rvey about the impact of JCL on an
Single incorrect letter, or operating environment. Our intent is to quantify the cost of JCL to an
installation. Fifty one installations responded. Generally, we asked
Adjacent letters transposed. what tools customers use to avoid JCL.
JES2 support of the I*JOBPARM ahd I*OUTPUT cards was changed
so that when the keyword search routine failed, a spelling correction [
routine was invoked. The change logs success or failure at correcting 1 JCL Avoidance Techniques
the keyword and the proposed corrected keyword. However, the cor- 1 [
rected keyword is not used to continue job processing, so that the 1 Method Percent 1 Lines Man
change is customer transparent. We chose this route because we were 1 Using 1 Needed Years
not certain that spelling correction was effective, and did not want to 1 1
commit ourselves to it before we knew the effect of the change. 1 TSO CLiSTs 71% 1 3000 1 1/4
1 1
Our results are mixed. As previous work with JECL cards had sug- 1 TSO Commands 40% 1 1800 3/4
gested, incorrect keywords do represent around a third of the errors. 1 1
However, the vast majority of those incorrect keywords do not reflect 1 SPF Panels 48% 1 1400 112
typographic errors. Instead, they represent conceptual errors: the 1 1
keyword is specified on the wrong control card rather than being miss- 1 VM/cMS EXECs 10% 1 500 1/4
pelled. 1 1
1 1 Class O-J- T
... 1
..
I~~ ~~
--------1 1 1
CO Automatic Spelling Correction 1 1 Training 55% 1 16 3
I I 1 1 1
Month JECL 1 Keyword 1 Fixed 1 lather Computers 21% 1
Errors 1 Errors 1 1 1 . 1
1 1 1 1 JCL Generators 13% 1
February, 1983 31 1 12 1 2 1 1 1
1 1 1 lather Solutions 40% 1
March, 1983 35 1 11 0 1 1 1
1 1 1 Migration Problems 19% 1
April, 1983 53 118 7 1 1 1
1 1 1 Favor Requirement 65% 1
I I
May, 1983 34 1 10 1
1 1
June, 1983 52 1 16 5 1 The survey tells us that TSO CLiSTs, Training, SPF Panels, and
1 1 TSO commands are popular ways to cope with JCL. Further, it tells us
1 1 that installations coping with JCL can easily invest a couple man-years
Total 205 1 67 15 1 in the problem. (I got man-years by dividing lines of code using the
1 1 empirical DOD formula: 10 lines of debugged (repeat, debugged) code
Percents 100% 1 33% 7% 1 a day, 240 work days in a year.) Finally, it tells us that people want
1 1 improvements to JCL.
I I
My conclusion is that a specific error message reporting the incor-
.rect keyword is appropriate, but that automatic spelling correction
needs more study.
August 15, 1983 August 15, 1983