Sticker price, engine power, and luxury features have long been selling points for vehicles. Over the years, however, MPG (miles per gallon) has increasingly played a role in swaying a car buyer to choose one model instead of another. For some drivers, the impetus for purchasing a fuel-efficient car has been to reduce pain at the pump. For others, it's also a matter of better environmental stewardship; less fuel consumption means a smaller carbon footprint.
Similarly, data center operators who formerly based hardware-purchasing decisions on sticker price, processing power, and features such as storage capacity are increasingly taking an interest in machines' overall energy efficiency -- that is, how much electricity a system consumes to perform a specific number of tasks. Again, the driving forces here are primarily economic. Energy efficiency means lower electricity bills, and the cost of powering a server is by no means trivial. Energy efficiency can also mean fewer machines and more floor space in the data center. Moreover, environmental concerns play a role.
[ Other groups have introduced energy-efficiency benchmarks geared at the data center, including the EPA, SNIA, and SPEC. | InfoWorld blogger Paul Venezia demands the banishment of power-hungry spinning disks from servers. ]
In an effort to provide companies with a means of measuring the energy efficiency of data center equipment, the Transaction Processing Performance Council (TPC) this week introduced a specification called the TPC-Energy spec. TPC-Energy is intended to complement the group's existing specs, designed to measure the raw performance of a data center system (including server, storage equipment, and database) simulating an OLTP workload, as well as the cost/performance stats of data center setup -- specifically, performance in the context of the setup's purchase cost.
Notably, TPC-Energy should not be confused with TPC-E. The latter is a stand-alone TPC benchmark that simulates the online-transaction processing workload of a brokerage firm and specifically refers to the number of Trade-Result transactions a server can sustain over a period of time.
Performance at a price
Measuring performance in the context of price is important. Consider a simple scenario: Two data center operators are running the exact same workloads -- same OS, same databases, same storage equipment, same everything except for the servers. Operator A has a high-end 64-socket machine loaded with the latest, fastest processor and loads of memory. Operator B has a single-socket machine with a midrange processor and modest memory. Presumably, the first setup will yield better performance -- but that data alone will be of limited interest to a data center operator on a budget and can't make purchases based solely on raw performance.