Blog Joyent

Har.

This seems about right, except that a true Solaris admin wouldn’t spill his coffee.

Just got around to listening, great show. I don’t want to be a hair splitter, but you missed an important distinction between “bandwidth” and “latency”, especially as it applies to grid computing. These days you can get huge bandwidth by filling a truck with hard drives: latency is the problem.

That may seem trite, but it’s a problem that doesn’t go away. In high speed computing the speed of light is woefully finite. When you ping someone, the speed of light costs you 1 millisecond per 150 kilometers (round trip). I’m in Boston, and that accounts for about a third of my ping to Google.

Worse still, in computation it might take a bit tens of thousands of clock cycles to hike across a single meter. It only makes sense to distribute tasks if their computation time is significantly longer than their transfer time. A grid that works over human sized distances will always have to wrestle with this problem.