A View from Will Knight
Making Computers Talk in their Sleep
A device called Somniloquy processes network traffic autonomously, allowing a computer’s CPU, hard disk, and display to be powered down.
While working on a story about routing Iinternet data based on electricity price fluctuations, I came across a clever idea for reducing the amount of power used by ordinary computers.
Researchers at Microsoft and UCSD created a network adapter dubbed Somniloquy (meaning to talk in one’s sleep) that can process network traffic autonomously, allowing a computer’s CPU, hard disk, display, and I/O buses to be powered down without losing connectivity.
As Bruce Maggs, VP of research at Akamai, points out in the story, energy-aware routing will only work if hardware uses significantly less power when idle. After we spoke, he sent me a link to the Somniloquy research project noting that it could help make existing hardware more power efficient.
The Microsoft-UCSD network interface (described in this paper) could take over many network-related tasks like bit torrent file-sharing, and managing a remote desktop connection and a VOIP account, allowing the connected machine to enter sleep mode without losing its network link.
The adapter consists of a gumstix module with a 200 MHz XScale processor, 64 MB of RAM, and a 2-GB SD memory card running embedded Linux. When the adapter detects that the connected machine has entered sleep mode, it copies over networking information and carries out simple communications on its behalf.
The researchers also show that the adapter can perform more complex tasks for its host. For example, they created a modified IM client capable of responding to network messages and waking the host computer when a proper message is received. They also developed a compact bit torrent client that continues to download a file while the host is in low-power mode.
It’s a smart idea and I wouldn’t be surprised to see such features in future desktop and laptop computers, especially as energy use becomes an increasing concern.