NetApp Inc. and a research team at the University of California Santa Cruz (UCSC) are working on a prototype of...
a storage system that gives new meaning to long-term archive.
The collaborative project incorporates commodity disks, Ethernet networking and memory chips originally designed for cell phones, pulling them together into hardware modules called tomes. Each tome combines MAID and solid-state disks and incorporates self-healing features. The goal is to develop storage systems that last for decades.
So far, the prototypes put together by researchers at the USCS labs have been made out of commodity external disks using iSCSI connections. The eventual design of the system's hardware modules, according to UCSC associate professor Ethan Miller, will consist of a single high-capacity SATA disk drive that can be spun down when not in use, a CPU, an Ethernet networking interface, flash storage and DRAM.
Another self-healing storage system
The plan is to combine the tomes into a grid system that would parallelize throughput for data protection and apply self-healing features to the overall system. While the self-healing storage aspect is similar to new systems from Atrato and Xiotech, Miller said his goal is not to heal individual disks or eliminate the need for disk replacement. "Those systems are offering a finite period of self-management and at the end of those three or five years, you'll probably need to get a new system," he said. "We're looking to make sure the data is preserved if disks fail. But if a disk breaks in a tome, we throw it out."
Instead, the focus for the Pergamum project is on eliminating data loss regardless of the hardware used. The system would pick one or more tomes to store archived data on. It would calculate parity information and spread the data out over 16 tomes, dedicating three of the tomes to parity. That means any three tomes in a given group could be lost and the data could be reconstructed. The system would also have RAID on disk, with two parity bits stored locally on each disk drive for every single bit of actual data. If a drive failed, the system would first try to reconstruct the data from the local bits in order to avoid multidrive rebuilds.
Miller said another goal for the Pergamum project is to lower power consumption. In addition to the ability to spin down disk drives when not in use and store the most frequently accessed data on low-power flash memory, each tome uses low-cost and low-power CPUs originally intended for mobile devices. The performance of these chips isn't terribly fast, but they can be bought in bulk for less than $40. The project is also working to incorporate the Power over Ethernet standard to deliver power to tomes through the networking interface.
Miller and graduate students Mark Storer and Kevin Greenan have led the project, but they have been joined by a technical advisor from NetApp's advanced technology group, Khaladar Voruganti. The group is a three-year-old business unit at NetApp meant to keep the company abreast of new technology on the market. It has procedures for turning science projects like this one into actual products, but NetApp is coy about whether this project will ever see the light of day as part of its portfolio.
"NetApp is reviewing the implications presented in the Pergamum paper and evaluating how we can apply these technologies to our archival data products," said NetApp in a statement emailed to SearchStorage.com. "We will continue to build systems that enable businesses to quickly read, browse and search archived data, while also reducing power consumption and offering superior performance."