OTTAWA—IT professionals may soon have a do-it-yourself alternative to commercial network-attached-storage products, according to a presentation by self-described Linux “kernel hacker” Ben LaHaise at the Ottawa Linux Symposium.
LaHaise revealed his vision to create a reliable storage solution using “commodity,” or off-the-shelf hardware, therefore dramatically reducing costs. The core of the project was to support large storage capacities, easy setup, and error detection.
The NAS project entails a Linux kernel driver, currently undergoing development, which passes storage requests directly to the hard drive. The scheme will provide redundancy by networking several servers over Ethernet.
The driver, dubbed NetMD, is not yet available but early versions are expected to be released in the next few weeks. The name NetMD comes from the names for Linuxs RAID and network block device implementations, md and nbd respectively.
Combining concepts from network attached storage and server clustering, an installation of NetMD could include as few as four servers. Starting with a frontend system for receiving requests, a minimum of three backend servers actually store the data. The frontend server transmits storage read and write requests via multicast UDP to all backend servers.
By using Ethernet as the connection method, NetMD is able to utilize an innovative method to detect data corruption.
LaHaises proceeding report states that “the main data integrity feature of NetMD comes from the Ethernet packets which are used to transmit read and write requests over the network. Many Ethernet cards are able to record the CRC of an incoming packet.”
Using the Ethernet hardware to control error correction frees the CPU for other tasks.
While saying he was not directly targeting vendors like Veritas and EMC, LaHaise noted that NetMD provided an alternative to some of their products. He called NetMD “small, stupid, and fast.”
Future plans for NetMD include documentation, RAID Level 5 support, and the ability to detect access patterns and relocate data accordingly for improved performance.
The software will be freely available when released within the next few months, and will run both on standard and embedded Linux systems. LeHaise said he was considering developing commercial hardware/software bundles, but he hadnt decided when, or how much, the products would cost.