AVI SILBERSCHATZ, Bell Laboratories, Murray Hill, NJ(avi@bell-labs.com)

STAN ZDONIK ET AL.1, Brown University, Providence, RI 02912(sbz@cs.brown.edu)

1. INTRODUCTION

The field of database systems research and development has been enormously successful over its 30-year history. It has led to a $10 billion industry with an installed base that touches virtually every major company in the world. It would be unthinkable to manage the large volume of valuable information that keeps corporations running with out support from commercial database management systems (DBMSs).

Today, the field of database research is largely defined by its previous successes, and much current research is aimed at increasing the functionality and performance of DBMSs. A DBMS is a very complex system incorporating a rich set of technologies. These technologies have been assembled in a way that is ideally suited for solving problems of large-scale data management in the corporate setting. However, a DBMS, like any large tool, places some requirements on the environment in which it is being used. The DBMS imposes some execution overhead, often requires a fairly high level of expertise to install and maintain, and only manages data that is in fairly specific file formats.

At the same time, the data that needs managing is changing radically and is being stored in places other than database systems (e.g., files). It is also obtained in large volumes from external sources, like sensors. While the trend of building more powerful database management systems has a place, there is also a need for data management in contexts that cannot cope with the overhead of a full-blown DBMS; many environments call for a much lighter-weight solution.

Sometimes, instead of using an existing tool in a new application, it is better to embed reusable components in order to make the resulting system more responsive. In some cases, it is the techniques that a tool embodies that are most reusable. We argue that this observation is true in many new data intensive applications. We would like to reuse database system components, but when that is inappropriate we must be willing to reuse our techniques and our experience in new ways.

If we look around at information that people use, we see many examples in which database systems are conspicuous by their absence. One of the most compelling examples is the World Wide Web. While it is true that DBMS vendors are making their products web enabled, their approach is to provide better web servers. This capability is only a very small step in the direction of managing the huge volume of nonstandard data that exists on the Web. It is doubtful that this move will cause the

hundreds of thousands of web sites to shift to the use of a full-featured database system whose target market is business data processing.

Other examples of applications that could benefit from data management techniques, but typically do not make heavy use of database products include personal information systems, news services, and scientific applications. In the case of personal information systems, one only has to think about the information found on the typical PC. Electronic mail is of great personal value to many users, but when messages are saved, they are most often stored in the file system. It would be extremely useful to have DBMS facilities such as indexing and querying available for use on email. While some support for a more organized approach to storage and retrieval of email is emerging (e.g., Lotus Notes), sophisticated querying is not well developed.

Other recent reports [Gray 1995; Silberschatz et al. 1991; Silberschatz et al. 1995] have charted the course of database research, and have done an excellent job prioritizing current research topics and delineating new influences with respect to their impact on the database system industry. This report takes a somewhat different tack. Our theme is that database research should be devoted to the problems of data management no matter where and in what form the data may be found. We should not be defined strictly by the current product space or by the commonly held notion that our job is to manage very large collections of structured records within a controlled environment. Instead, we should apply our skills to new data management environments that potentially require radically new software architectures.

ACM Computing Surveys, Vol. 28, No. 4, December 1996