Sun: Save energy, thread in multiples

The hardware maker believes multithreaded processors, with the ability to simultaneously compute many transactions, can lower a company's electricity bill.
Written by Eileen Yu, Senior Contributing Editor

Power fever may not have hit Asia yet, but there are growing concerns among companies, particularly data centers, in other parts of the world over rising utility bills.

To address the problem, chip and hardware makers including AMD, IBM and Intel have been focusing their attention on designing systems that consume less power.

Google's power bill, for instance, is the search giant's second highest expenditure after employee salaries, according to Marc Tremblay, Sun Fellow, vice president and chief architect of scalable systems group, Sun Microsystems. Google operates 250,000 servers across its data centers worldwide, handling nearly half of the world's Web searches.

Sun Microsystems believes the answer to the power problem lies in multithreading--a server's ability to simultaneously handle multiple tasks or transactions, said Tremblay, who was in Singapore Tuesday for a customer event.

Most systems today are underutilized, where only 25 percent of their computing power is exploited at one time. What that means is, when a transaction request is sent to a server, 75 percent of the system's memory lies dormant, or what Tremblay coins "memory latency". Through multithreading technology, a new transaction request is sent to the system so these inactive memory spaces can also be utilized, he said.

Through its Niagara chip, Sun has advocated the importance of multicore processors that have multithreading capabilities.

Each Niagara chip has eight cores, or processing engines. Each core is able to compute four transactions, dubbed threads, at the same time. Do the math and that will give each Niagara processor the capability of handling 32 transactions simultaneously.

Improving the chip's ability to simultaneously execute many transactions, such as a search request, will address the power problem, said Tremblay. It can also reduce real estate costs for data centers because more computing power can now be packed into a single system, he said.

One of 11 Sun Fellows worldwide, the highest accolade a Sun engineer can earn, Tremblay is also a co-architect of the first Sun UltraSparc processor.

Editorial standards