[Beowulf] Not quite Walmart, or, living without ECC?

Greg Lindahl lindahl at pbm.com
Mon Nov 26 13:53:45 PST 2007


On Mon, Nov 26, 2007 at 12:27:03PM -0800, David Mathog wrote:

> Anyway, this was the best test I could
> find for detecting the occasional gamma ray type data loss event. 

I always thought that the bit fade test was aimed at finding
manufacturing defects and the like.

> On the web there are references to an IBM study which found 1 bit
> error/256Mb/Month,

Note that the size of the memory chip plays into the rate, so you
can't directly compare with that number. (bigger chip = higher rate,
smaller fab process = lower rate, different fab process = different
rate, etc.)

-- greg




More information about the Beowulf mailing list