Tuesday, October 06, 2009

Google Finds DRAM Error Rates Higher Than Expected

For geeks only ZDNet reports DRAM error rates: Nightmare on DIMM street. "A two-and-a-half year study of DRAM on 10s of thousands Google servers found DIMM error rates are hundreds to thousands of times higher than thought — a mean of 3,751 correctable errors per DIMM per year."


Michael Critz said...

I'm going to note this the next time a nosey somebody snoops through my browser history.

"That wasn't what I was looking for. Did you know that Google's servers produce 10 mistakes PER DAY!"

Anonymous said...

I found the following comment fascinating....

For the frequency of soft errors. . . . IBM stated . . . that at sea level, a soft error event occurs once per month of constant use in a 128MB PC100 SDRAM module.

What's up with the sea level reference. Are these errors influenced by changes in barometric pressure or minute changes in gravitational forces?

I guess the chips in space are all messed up!


Howard said...

It wouldn't surprise me that there's an altitude difference, though it might have to do with the magnetic field of the earth.

I doubt the chips in space are off-the-shelf DRAM chips. I'm sure all the issues are spec'ed out.

Anonymous said...

Good point about the magnetic field. Even so, I think the changes in the magnetic field as a function of altitude are minute unless you are talking about big differences in the altitude (i.e., measured in miles not feet).