The human brain comes equipped with more transmitting neurons than it needs. When a neuron dies, the brain automatically lets other neurons take over. So why can't storage systems do the same thing when a disk drive fails -- without an administrator's help?
Moidin Mohiuddin and his research team at IBM's Almaden Research Center in San Jose are exploring that paradigm with Intelligent Bricks hardware and Kybos software. Together, they are creating an intelligent storage system architecture that will make data easier to manage, lower maintenance costs and reduce environmental impacts such as cooling, power, floor space and noise.
The basic unit of the scalable hardware is called a brick, an idea that was developed by IBM researcher Winfried Wilcke in 2001. Each brick contains a processor, some disks and an Ethernet switch chip. Bricks are placed next to one another and communicate with their neighbors without cables or connectors by using capacitive couplers on each brick face, forming a 3D, high-speed mesh network.
While each brick uses inexpensive hardware without internal redundancy, the entire 27-brick cube -- which is about the size of a small ottoman -- has a 26TB capacity and is highly reliable and interactive.
When used with the Kybos (Greek for "cube") software, the system senses that a brick or set of disk drives has failed and assigns new bricks to take over that function.
Software users will also be able to enter the performance requirements they desire, such as number of I/Os per second, maximum latency allowed or data-retrieval speed, and the software will configure the system to work at that level.
The key is providing a lot of redundancy, says Mohiuddin, senior manager of advanced storage subsystems at IBM. The prototype uses 25 percent more bricks than are actually needed to store the data, which allows for fail-over space. As a result, "you can wait four years before you do any servicing", because damaged disk drives don't need replacing quickly, he adds.
The bricks' small size cuts down on floor-space requirements, and they are cooled by water, not fans, reducing noise and power consumption.
One analyst gives this emerging technology mixed reviews. "It's not clear that [brick technology] will have a transformational impact on acquisition costs," says Gartner analyst Stanley Zaffos, who is familiar with brick and large-array hardware. "On the other hand, the software looks more interesting because it holds the promise of lowering TCO by creating an infrastructure that's more flexible, extensible and manageable."
With only a 10-member research team working on such a large project, Mohiuddin relied on partnerships with other IBM units to help with the prototype. IBM Germany, for instance, designed some of the circuit boards, while IBM's Watson Research Centre in New York helped to build capacitive couplers and assemble the bricks.
Today, the team must keep up with advances in disk drive technology. "Our first prototype had components that were available in 2001," Mohiuddin says. "So we want to redesign the whole system with newer components" that could increase data capacity on each brick from 1TB to 4TB.
By 2010, Mohiuddin says he expects to see the fruits of his research used in storage systems for medical images, simulation applications, Web sites and other storage-intensive applications.
Today, however, the hardware and software projects are progressing at separate paces. So Kybos might be available first for existing storage systems. "But the storage system would need high communication bandwidth because [Kybos] moves data around to meet your needs," according to Mohiuddin.
Together, Intelligent Bricks and Kybos create IBM's "grand vision" for future storage systems, he adds. "It's inexpensive, scalable and easy to administer."