Stepping up its campaign against the Linux operating system, Microsoft Corp. on Thursday released file and Web serving benchmark results that, it claims, show that Linux on the mainframe lags behind Windows 2003 on Intel Corp. systems in terms of performance for the money.
Microsoft ran the benchmark tests in response to what it believes were IBM's unsubstantiated claims about the benefits of running Linux on its mainframe hardware, according to Microsoft General Manager of Platform Strategy Martin Taylor. "IBM was making some claims and customers were trying to figure some things out and didn't have the facts in front of them," he said.
Taylor, who recently became MIcrosoft's Linux point man, has promised to focus on facts instead of emotional arguments when criticizing Linux and open source. Thursday's benchmark results appear to be a first effort in that direction.
It will not be the last. The company has also commissioned Forrester Research Inc. to prepare a study comparing Windows to Linux, according to a Microsoft spokeswoman who declined to comment on the specifics of the report.
The benchmark results discredit IBM's contention that a z900 mainframe processor running Linux could perform as well as three or four Intel processors running Windows at the same clock speed, Taylor said.
"One Linux image on one zSeries CPU performed at about the same as Windows NT 4 on one 900MHz Intel processor. Windows 2003 outperformed it. No question," he said.
However, Microsoft was unable to reveal at least one important fact relating to today's numbers: the source of IBM's alleged claim that a z900 processor running Linux could perform as well as three or four Intel processors running Windows. Microsoft maintains that the claim is being made to customers by IBM sales staff, but it was unable to provide any evidence of this. An IBM spokesman contacted about the issue knew nothing about this specific claim.
One industry observer welcomed Microsoft's benchmark numbers. "God help me for saying this, but I think Microsoft has made a genuine contribution here," said Rudy de Haas, an IT consultant based in Kitchener, Ontario, who has written about Linux's performance on the mainframe under the pseudonym Paul Murphy.
De Haas has criticized IBM in the past for not providing information on Linux's mainframe performance. The numbers released Thursday, though they did not come from an ideal source, were significant, he said. "There are several things you could use to raise questions about what (Microsoft) did, but the real bottom line is that no one else has done it."
The reason Linux mainframe benchmarks are so scarce is that they simply may not matter to customers, according to IDC analyst Dan Kusnetzky. "The bigger question is, why would people consider putting Linux on the mainframe?" he said. "Performance might be a secondary or tertiary issue. It may be an attempt to lower the cost of maintenance, and the raw performance may have been good enough for the task at hand."
Cost savings and ease of use were two of the driving factors behind the move to Linux on the mainframe by Pennsylvania retailer Boscov's Department Store LLC, according to the company's manager of technical support, Joe Poole. His company saved money on floor space, electricity, and even its disaster recovery subscription costs by moving a number of applications, including file and Web page serving, from Intel servers onto the mainframe.
"Your business applications run just fine on the mainframe. Your theoretical or forecasting applications, some of those things that rely on exponential mathematics, are sometimes better served on a different platform," he said. "But for our purposes, everything that we're doing at Boscov's is running just fine."
IBM has no plans to begin running Linux performance benchmarks for the mainframe, the IBM spokesman said. "We typically don't run industry standard performance benchmarks for any software on the mainframe," he said. "The value of the mainframe is based on its ability to securely run multiple applications on a single platform as opposed to purely seeking outstanding performance of one application on a single platform."
Microsoft's benchmarks were conducted by the Veritest division of Lionbridge Technologies Inc. under contract from Microsoft. They can be found here.