| By Maureen O'Gara |
Article Rating: |
|
| February 27, 2013 08:30 AM EST |
Reads: |
2,615 |
Baidu, which is often called the Google of China, has apparently won the race to become the first company to deploy servers based on ARM smartphone chips in a "large-scale" production environment in its data center.
It's using a custom version of Marvell's low-power 32-bit quad-core Armada XP CPU Server-on-a-Chip (SOC) in a personal cloud storage system called Baidu Cloud or Baidu Pan.
It's unclear whether any other kind of server is being used to deliver the service.
The boxes themselves, called "easy-to-integrate," were run up by an unidentified Asian ODM.
Marvell, like the other dozen or so ARM fabricators, is hoping to eventually challenge industry-standard x86 servers, a situation Intel would rather avoid. ARM has forced Intel to try to cut its inefficient power consumption.
Marvell figures the Baidu implementation validates its long-held belief that ARM technology will power everything from smartphones to large enterprise data centers.
The widgetry Baidu is using includes a Marvell storage controller and a 10Gb Ethernet switch integrated on the chip. The Marvell platform is designed to increase the amount of storage for conventional 2U chassis to 96TB using 24 2.5-inch hard disk drives and lower the TCO by 25% compared to x86-based server solutions.
Marvell also designed server management software compliant with the IPMI 2.0 specification for the storage server.
Marvell says the widgetry incorporates its Ethernet physical layer (PHY) transceivers, which offer the industry's lowest power dissipation, smallest form factor, highest performance and most advanced feature set. This PHY technology can cut power consumption to under 150 mWatts per port so network systems makers can decrease system cost by reducing both power supply and fan requirements.
Baidu should cut its power consumption in half.
The four 1.6 GHz ARM V7-based cores Baidu is using support 16GB DDR3 DRAM, which is supposed to lead to significant savings, probably the primary motivator behind the interest Facebook and Amazon share in the compute promise of ARM servers when they finally go 64-bit later this year. Storage doesn't need that kind of raw performance or 64-bits.
