HP breaks in new TPC power usage benchmark

The Transaction Processing Performance Council (TPC) has added a power efficiency metric, TPC-Energy, to its ongoing roundups of speedy transactional processing systems. Hewlett-Packard is the first participant to use the new TPC-Energy specification, submitting benchmark results of four of its systems.

"Energy is an increasing factor in the total cost of ownership" for many organizations, said Karl Huppler, chairman of the TPC. The new benchmark will allow an organization to judge how energy efficient systems are, as judged by watts per performance.

When combined with other TPC benchmarks, TPC-Energy can help customers weigh the trade-offs between price, performance and energy usage of any given transactional system, TPC says.

"One system may provide a higher performance level, but may not do as well in terms of watts per transaction," said Mike Nikolaiev, who is chairman of the TPC-Energy committee as well as a director of performance engineering at HP.

TPC-Energy, introduced in December, measures how many transactions a server can perform per watt. Transactions are measured by using one of three other TPC benchmarks for measuring server performance, TPC-E, TPC-C and TPC-H.

The TPC-E benchmark simulates a database-driven online transaction processing (OLTP) workload of a typical brokerage firm. An older benchmark, TPC-C also simulates an OLTP environment, for an order-entry workload. The third, TPC-H, simulates an ad-hoc querying, or decision support, workload.

HP ran four systems, each based on new ProLiant servers, through the test. For instance, under the TPC-E test, an HP ProLiant DL585 G7-based system was able to execute 1,400 transactions per second (TPS), using 6.72 watts. In contrast, another HP system, one based on the HP ProLiant DL580 G7, was able to execute 2,001 TPS in the TPC-E test with 5.84 watts.

Comparing the performance of the DL585 and DL580 using TPC-E with TPC-Energy benchmarks, one might conclude that the DL580 would be the best choice, both for TPS throughput as well as energy consumption. But the TPC also encourages users of the test to factor in system prices. In this case, the DL580 is more expensive than the DL585 when measured by dollar-to-TPS rating (US$347 versus $330).

The DL580-based system "was a little bit more expensive to get the higher efficiency," said Nikolaiev.

System power usage is measured by instrumenting each of the subsystems with a number of power analyzers, which feed data to software provided by TPC, called the TPC's Energy Measuring System, that collates the results.

The software can also break down energy usage per subsystem, showing how much the application server, database server, storage and other subsystems consume. Reporting on subsystem usage is an optional part of the test, Nikolaiev said.

As with other TPC benchmarks, vendors conduct the tests themselves, picking their own optimum configurations and submitting the results to TPC for consideration in its lists of top performers. Each set of results must be audited by a third party.

With HP as the first contributor, TPC is hoping that its other members will feel the peer pressure and submit the TPC-Energy results of their own systems. A nonprofit organization devoted to developing transaction processing and database benchmarks, TPC is funded by a swath of IT companies, including AMD, Cisco, Dell, HP, IBM, Intel, Microsoft, Oracle, Unisys and others.

TPC's work with TPC-Energy comes about at a time when the U.S. Environmental Protection Agency is ramping up its own guidelines for improving the energy efficiency of data centers.

Beyond TPC-Energy, TPC is working on a new set of benchmarks for data warehouse-styled workloads, called TPC-ETL, and virtualization workloads, TPC-Virtualization. It is also revising TPC-H.

Joab Jackson covers enterpise software and general technology breaking news for The IDG News Service. Follow Joab on Twitter at @Joab_Jackson. Joab's e-mail address is Joab_Jackson@idg.com