5 ms·
One bit per GB of RAM, not per GB of processed source code (most of your RAM is likely empty most of the time, or used for stuff where a bit flip is less likely
by jhugo 3y ago
One bit per GB of RAM, not per GB of processed source code (most of your RAM is likely empty most of the time, or used for stuff where a bit flip is less likely to be noticed). However, it still seems like a high estimate to me.
- pmontra 3y agoI have 32 GB of RAM. If I only use 1 GB of it it's still one bit flip in the RAM I'm using.
- jhugo 3y agoEh? The "cosmic ray" (or similar upset) is roughly equally likely to affect any bit in the 32GB, so the chance of it appearing in the 1GB that you're using right now is 32 times smaller than the chance of it happening anywhere in your RAM.
- zaphirplane 3y agoOperating systems keep files in ram In addition to memory allocation till they need the ram, so I think it should be mostly full all the time
- mrspuratic 3y agoUnused RAM ~= wasted RAM... Recently I had to troubleshoot a linux process that started segfaulting randomly. Having dealt with non-ECC bit flips before I took a copy of the binary (i.e. from the filesystem cache), dropped the caches, and compared and disassembled the binaries: a single bit flip resulted in an illegal instruction. Cost of investigating issues >>> cost of ECC
- jhugo 3y agoBut bit flips may not be noticed any time soon in the disk cache, and when they are later noticed, may be wrongly attributed to disk failure. It's a different failure mode than a bit flip in some source code that you are about to compile (which is far less likely to happen, due to that source code being a much "smaller target").
- geraldhh 3y agoit is my interpretation as well that most bit-errors occur in places that simply don't matter or are handled in userspace error-detection/correction (tcp-checksums, asset-checksums, encryption, etc)
- morelisp 3y agoAny RAM-caused bit flip that happens e.g. before the TCP checksum is as likely to happen after the TCP checksum, it's using about the same memory either way.
- geraldhh 3y agoleaving aside that a checksum doesn't catch every bit-change, that's still a 50% chance that it _will_ be detected.
- morelisp 3y agoI'm arguing against > most bit-errors... are handled in userspace error-detection If there really are a significant number of bad-RAM errors getting caught at the TCP layer, then there are also a significant number happening after that. There aren't, so there aren't.