First Name : Reza

Last Name : Taheri

Title :

Company : HP

Address : MS 44UG, 19447 Pruneridge Ave.

City : Cupertino

State : Ca

Mail Code : Telephone : (408)447-0013

FAX : (408)447-5958

E-mail : rtaheri@cup.hp.com

Cyberpaper : Eight years ago, at HPTS ’91, George Copeland gave a talk on whether the need for more OLTP systems, and for faster OLTP systems, was dying. If each person in the US generated 1 transaction per day, how many systems could we possibly need? His answer was that there are more transactions than that and our system do not process transactions with nearly as much speed and efficiency as benchmarks suggest. So, our jobs were safe for a while.

The question is relevant again. We have doubled performance every year for the past ten years, as measured by benchmark results. Today’s top TPC-C result is almost 200 times higher than the top mark in 4/93. That’s an annualized rate of 2.4X. Are our OLTP engines running out of things to do?

A colleague uses the changes in the PC market to further argue for this point. For the first 15 years or so of the PC era, the sweet spot of the market —in terms of price— remained about the same, but the PCs kept getting faster. The systems were never quite fast enough and we all wanted the next generation. Then, it seems the products caught up with our needs. The past 2-3 years has not seen as much increase in the processing power, but a great reduction in the price of PCs. Will the same phenomenon occur in the OLTP server market?

Well, no! For one thing, don’t believe everything you read in benchmark results. Systems have definitely gotten faster, but not to the tune of 2.4X per year. (I can expand on this if need be.) But more importantly, the demand is about to explode. E-commerce will generate transactions at a rate we haven’t dreamed of. Some of the stuff you read in the futuristic ads will really happen. And the transactions will become more complex with Java code and other interpreted content, authorization, authentication, and much more frequent interactions among multiple, dissimilar systems. Some of the classic transaction processing topics, e.g., maintaining ACID properties among many dissimilar systems connected via unreliable links, will start finding wider application in the real world. We’ll continue to have job security for a while.