IBMs p570 Server Gets Big Boost from Power6

 
 
By Wayne Rash  |  Posted 2007-10-25 Email Print this article Print
 
 
 
 
 
 
 

eWEEK Labs: The first server designed specifically to use IBM's Power6 is robust yet energy-efficient, but it's the processor that's the star of the show.

AUSTIN, Texas–IBMs p570 is the first in a new line of servers designed specifically to take advantage of the companys new Power6 dual-core processor. While the p570 has grown from a line of earlier p-series enterprise machines, it offers a number of improvements and enhancements, some enabled by the Power6.

The processor, meanwhile, stems from earlier Power-series processors, but it is a significant step forward in processor capacity and reliability.
I was invited to IBMs Austin labs to get a chance to evaluate the p570 and the Power6 that drives it. The p570 is a modular, dual-processor server that can be expanded to as many as 16 processor cores. (Note that each Power6 processor has dual cores, so IBM counts these as two processors each.)
Each server uses blade-like modules for functions ranging from carrying the processor cards to handling storage. In addition, as many as four server chassis can be linked to effectively function as a single unit. Some of the p570s hardware features, including its Ethernet adapters, have been designed for virtual operations. Dual-port Gigabit Ethernet cards come standard on the p570, and a quad-port or 10 Gigabit Ethernet adapter can be added. Each p570 can support as many as six SAS drives. Each server has three PCIe and two PCI-X expansion slots. The p570 chassis supports redundant power supplies, and the server supports up to 192GB of system memory.
Click here to read the Labs take on the Power6 and quantum mechanics. IBM claims a number of benchmark firsts with its 4.7GHz Power6-based machine. While I wasnt able to confirm any of them (eWEEK Labs wasnt given access to the hardware for testing), the servers specifications are certainly impressive. According to IBM, the new p570, which runs only Unix or Linux, uses the same amount of electricity as earlier p-series computers that only ran half as fast. Company officials say the p570 can handle data at speeds up to 300G per second. IBM also claims that the server will be energy-efficient. According to IBM, the machine can save up to $100,000 in energy costs compared with SunFire v890 servers from Sun Microsystems. Still, this is an enterprise-class machine with enterprise-class power needs, running at 240 volts. The Real Star As compelling as the p570 is as a whole, the real star of the show is the Power6 processor. This dual-core processor includes decimal floating-point processing and a 65-nanometer design. The chip runs internally at 1 volt, although it has active voltage control that will change the voltage supplied to parts of the processor as needed. To make the processor run effectively at 4.7GHz, IBM had to make some dramatic changes. Some of these changes, such as revising the flow of the instruction pipeline, make the processor more efficient. Others, such as adding the ability to locate cached instructions quickly, involve innovative ways of storing information and then finding it again. And one big change—the ability of the processor to monitor itself for errors in executing instructions and then fix the problem—are unprecedented in small computers. According to company engineers, technology formerly available only on mainframe computers was brought into the processor design. In short, IBM has basically instrumented everything inside the processor. The Power6 can tell when an instruction isnt being handled properly and re-do the instruction. In addition, it can keep an eye on heat, oscillator performance, and controller and register operations, and it can monitor its own critical path. The operations checking, which happens every clock cycle, include dynamic bitline repair, failover for oscillators and controllers, cache recovery and, if needed, alternate processor recovery. This means that if something goes wrong that can be corrected, the processor corrects it. And, if theres a hardware failure that cant be fixed, a new processor or component can be brought in automatically. Because most of the error checking and recovery happens at the chip level, software running on the server never knows that there was a problem. The server keeps on running without a hitch. And even when theres a non-recoverable hardware error, the processor can bring another core into operation and keep running. Perhaps the biggest change that software will notice is the ability to handle floating point decimal math using the IEEE 754r format. This spec was developed by IBM and released to the open-source and GNU communities. The spec is supported by Unix and Linux, but is currently available in hardware only from IBM. IBM officials also claim that server downtime is extremely rare due to factors ranging from the ability of the servers processor to bring spare components online as needed to the support for decimal floating point. Technical Analyst Wayne Rash can be reached at wayne_rash@ziffdavis.com. Check out eWEEK.coms for the latest news, reviews and analysis on enterprise and small business storage hardware and software.
 
 
 
 
Wayne Rash Wayne Rash is a Senior Analyst for eWEEK Labs and runs the magazine's Washington Bureau. Prior to joining eWEEK as a Senior Writer on wireless technology, he was a Senior Contributing Editor and previously a Senior Analyst in the InfoWorld Test Center. He was also a reviewer for Federal Computer Week and Information Security Magazine. Previously, he ran the reviews and events departments at CMP's InternetWeek.

He is a retired naval officer, a former principal at American Management Systems and a long-time columnist for Byte Magazine. He is a regular contributor to Plane & Pilot Magazine and The Washington Post.
 
 
 
 
 
 
 

Submit a Comment

Loading Comments...

 
Manage your Newsletters: Login   Register My Newsletters























 
 
 
 
 
 
 
 
 
 
 
Rocket Fuel