In-Depth

pSeries Grabs Performance Crown

A pSeries system anchored by IBM’s new Power5 processor obliterates the competition in a standard industry benchmark

• By Stephen Swoyer
• 11/30/2004

It’s a truism that you should take all benchmark tests with a grain of salt. Even so, a new result from IBM Corp. might be the exception to that rule.

In a coming-out party of sorts for its new Power5 processor, Big Blue posted a record benchmark in the Transaction Processing Performance Council’s (TPC) standard TPC-C test that far outstripped all previous marks.

A 64-way pSeries p595 system from IBM notched a score of 3,210,540 tpmC on the popular industry benchmark. That’s nearly three times better than the number two score—1,184,893 tpmC—achieved by a 64-way cluster of Integrity 5670 servers from Hewlett-Packard Co.

What’s more, IBM’s 64-way Power5 system more than tripled the best non-clustered result, a 64-way HP Integrity SuperDome system.

Benchmarks, of course, are a tendentious thing. “You wouldn’t believe what companies do with benchmarks,” says industry veteran Mike Schiff, a principal with data warehousing consultancy MAS Strategies who has also logged time with Oracle Corp., Software AG, and other vendors. “Oracle once had something called discrete transactions that are built into the database just for the purpose of enhancing its performance in benchmarks.”

In a teleconference last week with reporters, Adalio Sanchez, general manager of IBM’s pSeries server line, defended the TPC-C benchmark as the “gold standard” in business computing. In fact, Sanchez argued, the TPC-C metric is prized by many customers as a reasonably accurate indicator of system performance in real-world situations.

IBM and HP have in the past played a game of leapfrog in the TPC-C benchmark (see http://www.esj.com/news/article.aspx?EditorialsID=647). HP seemed to more or less permanently take the TPC-C lead when it posted the current number two score late last year—but that was before IBM delivered Power5. In this case, Sanchez says, Big Blue hasn’t simply grabbed a temporary lead in the TPC-C test. “[This is] a performance delta that is well beyond what we’ve seen in the recent past,” even suggesting that Power5 ““is really changing the game here. In fact, [the game] may be over.”

Sanchez says IBM’s record TPC-C result is a validation of its Power strategy, which combines an aggressive processor roadmap with a complementary chipset architecture and other integrated technologies. “We made a big bet on microprocessor and systems technology integration. At the time, the conventional approach was to crank up the frequency and shrink the chip to drive systems performance,” he says. “However, it became clear to us … that the way technologies are integrated together would be the fundamental approach to drive systems performance and to keep pace with Moore’s law.”

The p595’s TPC-C record benchmark also highlights the compelling price/performance of IBM’s Power-driven pSeries systems, Sanchez said. Big Blue’s number one result was cheaper—by almost 6 percent—than HP’s second-place (clustered) system, and the p595’s performance was significantly less expensive—by about 33 percent—than HP’s best non-clustered result.

All benchmarks should be taken with a grain of salt, says MAS Strategies Schiff, but IBM’s Power5 is clearly a performer to be reckoned with. “This sort of establishes at least one sample point showing the performance of this architecture,” he comments. “You can only tweak these [benchmarks] within certain limits. So you can optimize for the benchmark, but you’re still going to have to have the inherent power there. Your Corvette is probably always going to beat your Ford Explorer in a drag race.”