[3B2-9] man2013020069.

3d 17/6/013 9:41 Page 69

Anecdotes

SQL/DS: IBM’s First RDBMS

Hershel Harris and Bert Nicol
Editor: Craig Partridge

In the late 1970s, IBM software labs were aligned with Thus, introducing any new database offering based
the IBM hardware families. The decisions to commer- on System R required careful positioning. An MVS
cialize the relational database prototype called System product would require a significant rewrite from the
R, which had been developed during the 1970s at the VM version of System R. It would also need to be posi-
IBM Research facility in San Jose, California, were tioned relative to the already successful IMS product
made based on a hardware family business case. The to help customers understand which to use. Finally,
Endicott Lab, supporting the small- to mid-sized main- it would need extensive quality and performance test-
frame environments running VM and VSE, had the ing along with lengthy field trails to not introduce
skills and the competitive pressure to launch the rela- any weakness into the fastest growing mainframe
tional database management system (RDBMS) commer- platform. An MVS version of System R would not be
cialization project in 1979, and it delivered SQL/DS two produced quickly.
years later. This article traces how SQL/DS, running on VSE, on the other hand, was used as a transaction-
VSE and then on VM, became IBM’s first commercial processing environment by thousands of mid-sized cus-
relational database in 1982, over a year before the avail- tomers, but without the widespread use of databases.
ability of DB2 running on MVS. Mission-critical, real-time VSE applications usually
accessed data files through CICS or with non-IBM trans-
Focused on Hardware action-processing offerings. There was limited use of
Until the mid-1980s, IBM was largely organized DL/I and other nonrelational databases. VM was used
around its hardware families of mainframe, midrange, primarily as a development environment or hosting en-
and personal computers. System software was devel- vironment for VSE. A commercial version of System R
oped by different labs within these hardware-based could be introduced into VSE or VM more quickly
organizations to exploit the capabilities of each hard- than into MVS, with less development effort and prod-
ware platform and to provide vertically optimized uct positioning.
solutions that would drive their use. The needed soft-
ware was mostly hardware specific, such as operating Internal Competition in IBM
systems, device controllers, and file systems. When The IBM Santa Teresa Lab (STL, now called the IBM Sil-
relational database technology emerged from IBM’s icon Valley Lab) had responsibility for software running
San Jose Research facility through the System R re- on MVS, including IMS. After evaluating the possibility
search project, IBM certainly didn’t see it as a new of integrating relational capability from System R into
software business category, but primarily as a way to IMS, they decided to keep them separate and develop
further increase demand for hardware. Around 1979, plans to build a new RDBMS on all major mainframe
the decisions to exploit this new opportunity were platforms (MVS, VM, and VSE). However, their budget
independently evaluated by the hardware groups, and skills would limit them to building only the MVS
for each operating environment, based on strategic version.
initiatives and market-specific pressures. Around the same time, the IBM lab in Endicott,
In 1979, MVS was the premier IBM operating envi- New York, which developed software for smaller
ronment, with a reputation for high performance, mainframes running VM and VSE, had just canceled
quality, and high availability. These attributes were a project to merge the VM and VSE operating systems
strengthened by a lengthy field trial process for new and had more budget and skilled programmers avail-
releases. IMS, the popular online hierarchical data- able than they had projects. So, the decision was
base and transaction-processing system on MVS, made to focus the development of the VM and VSE
along with nonrelational DBMS products from inde- RDBMS products in Endicott. This decision was also
pendent software vendors (ISVs) were successfully deliberate. Senior IBM executives saw the value in
driving mainframe growth, and the platform had lit- having a lab other than STL developing the mid-
tle competition. Customers desiring compatibility range RDBMS products to help motivate Santa Teresa
and consistency between database usage on MVS to hold to a competitive schedule; it also responded to
and VSE chose the Customer Information Control the threat presented by Oracle’s introduction of an
System (CICS) as a transaction manager, which RDBMS focused on mid-range competitive hardware
could operate with IMS databases on MVS or DL/I (DEC PDP 11 and then the DEC VAX/VMS platform).
databases on VSE (or even the database systems Endicott was challenged to deliver a RDBMS software
from the ISVs). product within two years.

IEEE Annals of the History of Computing Published by the IEEE Computer Society 1058-6180/13/$31.00  2013 IEEE
c 69
 [3B2-9] man2013020069.3d 17/6/013 9:41 Page 70

Anecdotes

Turning System R into SQL/DS on VSE next several years, new releases of SQL/DS
To meet the challenge of delivery within two version 1 enhanced the performance, us-
years, the team in Endicott chose not to rein- ability, connectivity, operating system ex-
vent the wheel. They didn’t replan the release ploitation, and SQL support of the IBM
content or redesign major portions of the RDBMS on VM and VSE.
program. They felt that if they just focused
on ‘‘productizing’’ System R they could re- SQL/DS Evolves
lease in two years or less, which they were Although IBM was still hardware oriented, a
able to do. shift was starting to occur in the mid-1980s
The part of System R that handles the in- where software development labs, unaffili-
terpretation of SQL and optimized path selec- ated with an IBM hardware division, were
tion is called the Relational Data System building IBM software products. One such fa-
(RDS), and it was about 38,000 lines of PL/I cility, the Toronto Development Lab, had re-
code and about 9,000 lines written in Assem- sponsibility for compilers and software
bler. The Research Storage System (RSS), development utilities on a number of IBM
which handled file access, indexes, and lock- platforms. As the Endicott team focused on
ing, was around 35,000 lines of code written developing components for the VM operat-
in PL/S. Development standards for product ing environment, such as the shared file sys-
offerings at that time required that the code tem (SFS), development responsibility for
be written in PL/S (or Assembler). Therefore, SQL/DS moved to the IBM Toronto Lab in
the RDS had to be rewritten from the some- 1985. There they built SQL/DS version 2 re-
what similar PL/I. lease 1 and had responsibility for all subse-
Endicott decided that it would release its quent releases. Version 2 releases were made
first product on VSE, even though the under- available in 1987 and 1989 and extended
lying System R prototype was built on VM Endicott’s work by including relational
(and so a VM product might seem easier to enhancements such as referential integrity
construct). Endicott did this because VSE and new data types. Performance and reli-
was more often used as a transaction- ability were improved with the incorporation
processing environment and had much of a newly architected plan generator and
wider commercial use than VM. In addition, database switching, providing additional
a production version of System R depended connectivity for VM customers.
on shared memory, which VSE supported In the early 1980s, the mid-range AS/400
and VM did not. Finally, to ensure that product line, developed at the IBM lab in
SQL/DS met the transaction-processing Rochester, Minnesota, used the RSS technol-
needs of VSE’s commercial customers, Endi- ogy within its integrated file system, and in
cott chose CICS as the transaction manager 1988, Rochester released SQL support. In
and carefully modeled the SQL/DS system 1989, the IBM lab in Austin, Texas released
to ensure it would support competitive trans- an extended edition of OS/2 for the IBM
action rates. PC, which included relational database ca-
SQL was then reviewed and updated to en- pability. By that time, IBM had SQL data-
sure that every operating system environ- bases on five different platforms, built by
ment could use the same version and that it four different teams, each with their own
could be successfully embedded within the levels of SQL support and different opera-
various programming languages used to tional characteristics, making application
write SQL-based applications. This process portability difficult for both third-party
was the start of the cross-company SQL coun- software developers and customers.
cils, which drove SQL as one of the early IBM had just reorganized to better provide
open standards in computing. cross-platform software solutions and de-
Finally, to continue supporting customer cided to move the OS/2 database manager
use of DL/I on VSE, a bridge module was to the Toronto Lab. This allowed the same
planned and developed to support interoper- group that was supporting and enhancing
ability with SQL/DS for data management. SQL/DS to build a common version of an
SQL/DS release 1 was announced in IBM RDBMS on OS/2 on AIX and, for the
1981 and made available in early 1982 on first time, on non-IBM operating system plat-
VSE. By 1983, VM provided the shared- forms such as Microsoft Windows, Sun
memory mechanisms needed for SQL/DS Solaris and HP/UX. This was a significant
to operate, and SQL/DS R2 was then step in moving IBM to becoming a software
released for the VM platform. Over the and solutions company, and it allowed for

70 IEEE Annals of the History of Computing

[3B2-9] man2013020069.3d 17/6/013 9:41 Page 71

IBM’s subsequent delivery of an RDBMS on

many Linux platforms.
SQL/DS version 3 was delivered in five The harvesting of IBM’s
releases between 1990 and 1995. The first
two releases focused on SQL compatibility System R research
across platforms and FIPS SQL compliance.
Subsequent releases added significant inter- technology within and
operability improvements through support
of the Distributed Relational Data Architec- outside IBM created
ture (DRDA) standard, while continuing
platform exploitation with improved a multibillion dollar
extended memory and virtual machine sup-
port in VM and VSE. relational database
Sharing the DB2 Name industry.
The naming of IBM’s various RDBMS prod-
ucts was simplified when the new OS/2 and
AIX RDBMS versions were released as DB2, with its installed IMS and DL/I customer
the name already in use for the MVS base.
RDBMS product. Over time, the rest of the Today, all but two of the early nonrela-
RDBMS family aligned to use the same tional vendors are gone (although their prod-
name, and SQL/DS was released as DB2 for ucts are still in use), and IBM continues to
VSE and VM V5 to align with DB2 for MVS support and develop nonrelational and rela-
V5 in 1997. These versions delivered intero- tional technology as well as many new
perability and tool enhancements, including ‘‘post-relational’’ technologies that have
support for the DB2 Control Center, which arrived over the last 30 years. Few technology
provided a common interface for the admin- transfers have had the impact of relational
istration and management of DB2 on all plat- databases.
forms. Subsequent releases were delivered as
DB2 for VSE and VM version 6 and the cur- Hershel Harris worked in the IBM Software
rent version 7 to improve and extend intero- Group in the Toronto Lab and at the IBM Software
perability and better exploit new operating Headquarters in Somers, New York, starting in
system capabilities. 1978. He was part of the SQL/DS and DB2 teams
By the mid-1990s, as its name was from 1987 through 2001. As chief technology officer
of Georgian Partners since 2009, Hershel has helped
changed to DB2, more than 7,000 mainframe
software companies develop products and leverage
customers were using SQL/DS and over
their data using applied analytics. Contact him at
10,000 were using DB2 on MVS. [email protected].

The Value of RDBMS Technology Transfer Bert Nicol was part of the team that transferred
The harvesting of IBM’s System R research SQL/DS from the IBM Endicott Lab to the IBM To-
technology within and outside IBM created ronto Lab in 1985. He worked on the SQL/DS and
a multibillion dollar relational database in- DB2 teams through 2007. Contact him at nicolgrp@
dustry. It was arguably the most successful cogeco.ca.
demonstration of the commercial impact
of software research. IBM was one of many
companies selling nonrelational databases,
but it was the only one that then fully adopted
the relational model and navigated the
risks of introducing an offering competitive

April–June 2013 71