Total IBM/Red Hat hardware and software costs were $1.99 per transaction, which is 22% cheaper than the next less expensive hardware/software combination, (IBM/AIX,) at $2.81 per transaction and 300% less than the most expensive combination, HP-UX/HP Integrity Superdome, at $8.33 per transaction to exceed the million-transaction threshold. The TPC-C price/performance results measure online transaction processing rates and cost.
Red Hat's high ranking in the million-plus category didn't extend to the top 10 list on price/performance overall, however, because there are many less expensive competitors that can't scale to the same volume.
Red Hat also processed more than 100,000 operations per second per Java Virtual Machine for both bare-metal and virtual instances. Comparisons are difficult because results vary on different virtual machines; nonetheless, Red Hat operated with less overhead, which translates into less performance loss because of virtualization, according to Red Hat.
The test results show that Linux running on x86 chips can scale to high volume with commodity hardware at the most economical price-to-performance ratio, according to Red Hat marketing manager Andy Cathrow. No Windows-based system performed as well, and Unix systems with faster performance cost 50% more, he said.Democratizing transaction processing
"This puts transaction processing within reach of every customer," Cathrow said. "They don't need Unix for their highest workloads."
John Shakshober, Red Hat's director of performance engineering, noted that Red Hat has a precedent for breaking records and surpassing rivals. Several years ago, Red Hat was the first to break the million mark with clustered servers; now it has broken the mark with a single server. Not only is it the first million-transaction processor running standard x86 architecture, but it's on the new Intel-based Xeon 7400 with a six-core CPU, he said. The tests, which were completed on Aug. 19, often take months to perform, he said.
Hardware vendors typically publish only those results they know will be favorable to their products, Cathrow added. IBM chose Red Hat because IBM knew it could scale to maximum performance on its DB2 database, he said.
Joe Clabby, the president of Falmouth, Maine-based Clabby Analytics, said TPC's benchmarks are a good measure of transaction processing and, in turn, computing power, and show which vendors are leading at a given point in time.
These results are good news for buyers because it confirms that Linux can indeed perform high transaction rates at low cost, Clabby said. And by affirming Red Hat as the Linux TPC-C leader and validating IBM's strong performance as well, that's good news for Red Hat and IBM he said.