Contents Under Pressure 2

An examination of the world's largest and most heavily used databases unveils a wealth of information about current scalability challenges.
Regardless of usage, symmetric multiprocessing (SMP) prevailed as the leading architecture for databases with less than 2 TB of data in both programs. Among DSS databases, SMP systems strengthened their position in terms of midsize database platforms. Cluster systems have established a noticeable presence at the largest DSS database sites reporting in our program, penetrating territory previously outside their comfort zone.

You can see in Figures 5 and 6 that massively parallel processing (MPP) systems, whose demise has been predicted periodically by analysts, competing vendors, and some in the media, remain one of the principal architectural choices for the largest DSS databases. Their ongoing presence brings to mind Mark Twain's famous saying, "Reports of my death are greatly exaggerated." Although MPP platforms no longer play in the OLTP arena, it is abundantly clear from the survey data that MPP continues to enjoy vibrant growth for high-end decision support.

Among transaction-processing databases, SMP systems are used extensively, regardless of database size. They are the leading choice for the largest OLTP databases, weighing in at more than 5 TB, as well as for emerging large databases, which we measure as less than 2 TB. Cluster systems are the architecture of choice for midsize OLTP databases of 2 to 5 TB.

Overall, use of SMP and cluster architectures for both decision support and transaction processing grew significantly between 2001 and 2003. Uniprocessing, which barely had a presence in 2001 and even then was used only for transaction processing, has all but disappeared as a choice for large database architectures.

Scaling Ever Larger

Putting all this together, just what do we foresee for large databases in the near future? Most assuredly, they will become bigger, hopefully better, and definitely more powerful. In the next couple of years, we expect respondents to our survey with DSS databases to have 100- to 200-TB databases at the inquisitive fingertips of a wide range of users. We could see processing in the range of 1,500 concurrent in-flight DSS queries. The accompanying sidebar interview with Ken Collins, director of information management at, testifies to increased business-critical demand for enormous DSS databases.

Meanwhile, transaction processing will reach 8,000 tps against 25 TB of operational data. While these DSS and OLTP numbers may have IT executives shaking their heads, they are supported by the projections of our survey respondents. Our program will soon be gathering and compiling data for 2004, and there is little doubt that we will see databases bursting out of their confines, making scalability of every sort a critical issue for IT managers going forward.

Authors' Note: The Winter Corp. 2003 TopTen Program was sponsored by HP, Microsoft, Oracle, Sybase, and Teradata. Sponsorship brought about promotional benefits but did not impact the program findings whatsoever. The authors gratefully acknowledge the efforts of Vic Goldberg and Rick Burns, who contributed to the collection, management, and analysis of the data employed in the TopTen Program.

RICHARD WINTER is president of Winter Corp. He leads an international consulting practice on the technology and implementation of large databases.

KATHY AUERBACH is research program director of Winter Corp., where she has managed the TopTen Research Program and related research programs for the past seven years.

For more information about the survey reported in this article, see: