Three computer scientists from Google Inc. have published a study on â€œFailure Trends in a Large Disk Drive Population,â€ a thesis they presented recently at the 5th USENIX Conference on File and Storage Technologies (FAST’07), February 13-16, 2007 in San Jose, CA.
Eduardo Pinheiro, Wolf-Dietrich Weber and Luis Andre Barroso culled their data by examining the population of hard drives under deployment within Googleâ€™s computing infrastructure.
Itâ€™s an interesting academic study, especially if youâ€™re a geek too. But if youâ€™re only half a geek who gets sleepy reading â€œthesicalâ€ language, Iâ€™ll tell you what it says:
1) 90% of all new information produced in the world is being stored on magnetic media, most of it on hard disk drives;
2) Disk drives are generally very reliable but are also very complex components;
3) They canâ€™t really tell exactly when a hard drive is going to fail or why. Hey, things just happen, you know.
Which reminds us of that Murphy guy, â€œIf a hard drive can fail, it will fail at the worst possible time for you,â€ or something like that. So back up, kid, always back up if that data is really important (Oh, you know that already).
[Paper: Google Labs, PDF 242KB]