Coercing DBMSs to cooperate - DBMS: Distributed Databases - includes related articles titled 'IBM's distributed database direction,' military's Cals provides distributed DBMS model,' and 'importance of optimization'

Software Magazine, Nov, 1989 by Barbara Bochenski, Mike Bucken

COERCING DBMSs TO COOPERATE

Single-site DBMSs will be an antique curiosity within 10 years, predicts Michael Stonebraker of the University of California in Berkeley.

"In my opinion, distributed database systems will have a profound impact on DBMS clients over the next decade. In short, I expect everybody will be using a distributed DBMS within the next 10 years."

Stonebraker was involved with one of the earliest and most well-known distributed database (DDB) prototypes: Distributed Ingres. He says the success of that project--started in 1977--was constrained by the lack of available networking software.

Networking software is an important part of a distributed database environment. Charles Bontempo, a consulting staff member with IBM Systems Research, Shornwood, N.Y., defines a distributed database as a system involving multiple computer sites connected with a communications network.

Each site has its own database processed by its own DBMS, its own CPU and its own terminals. Each site also performs its own recovery management and logging.

Bontempo, speaking at a recent DB2 Users Group meeting, said each site is highly autonomous and independent, but cooperative.

The last 10 years have brought much progress to the distributed database scene. Increased networking capabilities have contributed significantly to the potential of distributed databases. Software solutions for major problems like crash recovery are increasingly available in the marketplace.

The increased intelligence and power of workstations, and the growing use of relational database management systems (RDBMS) and SQL have also contributed to the impetus. Relational databases and SQL have unique qualities that facilitate the use of distributed architectures.

Has the progress been enough? Are the commercial distributed database management systems (DDBMS) of industrial strength? Or are they just for small, specialized applications that can run on a local area network? Can distributed DBMS meet a multinational corporation's high performance needs on a wide area network? Do they fill a real need or is it just too soon for many enterprises to move toward distributed DBMSs?

Analysts differ about how ready industry is (and how ready the vendors are). Some say that enterprises cannot manage the network problems of a distributed database if a wide area network (WAN) is involved. Others thinks think many companies are ready to tackle the problems, but that vendor offerings of two-phase commit are not yet adequate.

Some believe that as IBM provides distributed DBMS capabilities with the upcoming shipments of DB2 version 2.2, many companies will implement the technology.

UNRESOLVED PROBLEMS

"The state of the technology today," says Kent Laver, a senior marketing specialist with Cognos in Toronto, "is such that we can very efficiently handle distributed transactions on a LAN. However, the problems of distributed DBMSs on WANs have largely been unsolved."

Cognos markets StarBase, a distributed database for LAN implementations on DEC, Hewlett-Packard and Data General systems. The database engine for StarBase is licensed from Interbase Software Corp., Bedford, Mass. Cognos' StarNet accesses databases via DECnet, HP's AdvanceNet and TCP/IP.

"The problem of distributed databases on LANs vs. WANs is one of management--system management in particular," says Laver. "Many corporations are in a position where they can manage a local area network. They have been doing it for mail systems for a while.

"But I think the problems involved in wide area networking--where you're interested in response times--have largely been unsolved. If they have been solved, they have probably been dealt with by individual corporations who have been able to put their own structure in place as opposed to buying something commercial."

Gordon Smith, senior product manager with Oracle Corporation, Belmont, Calif., basically agrees, though he looks at it a little differently.

"The actual technology is basically the same," says Smith, "whether it's on LANs or WANs. The database doesn't necessarily do things terribly differently. It would do some optimization perhaps differently in the two environments.

"It's more of a performance issue. There are many situations where you need to join lots of tables between two distributed sites. That will just take a lot longer on a WAN than a LAN. It's a cost-performance issue," Smith said.

ANTICIPATING NEW VERSION OF DB2

Roman Rozenberg, director of database research at Knauer Consulting in New York, believes that industry is definitely ready for distributed databases. "Some companies are just waiting for DB2 version 2.2 to be shipped," says Rozenberg.

A large bank with worldwide offices has started to implement an international network in anticipation of the shipment of DB2 version 2.2, says Rozenberg. When the product is shipped, they plan to install it and build their distributed architecture.

Several large companies with similar plans did not want to be identified or provide details because they believe their distributed database plans are of strategic importance.