Thanks for these kind and encouraging words. However, I'm not striving for a big audience. This weblog (and the RSS reader of Radio) are for me a means to discover new ideas and jot them down. This process often makes the idea to stick in my mind, so that I can recover it later. I have found that almost every week I discover something new through this weblog. The discovery may lead to a newspaper column, an article, or a message to a newly discovered colleague. And of course, I want to keep on writing something in English every day.

Well, the issue is becoming moot, because the VT G5 cluster is moving to Xserve G5s within four months. These systems have ECC memory. (And they require less space and consume less electricity.)

But in the original configuration this was certainly a problem, although perhaps a minor one. Using error-correcting memory in long computations is recommended, but there are of course other sources of error besides the memory. In well-executed computational science the computations are repeated enough times and independently from each run so that the external sources of error are eliminated (at least in theory). Of course, the question of modeling errors, method errors, and bugs in the code still remains.