Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!asuvax!ncar!ico!rcd From: rcd@ico.isc.com (Dick Dunn) Newsgroups: comp.unix.sysv386 Subject: Re: Automatic bad sector mapping Message-ID: <1991Jun10.230223.10316@ico.isc.com> Date: 10 Jun 91 23:02:23 GMT References: <767@dumbcat.sf.ca.us> <1991Jun10.025527.10161@jwt.UUCP> Organization: Interactive Systems Corporation, Boulder, CO Lines: 30 john@jwt.UUCP (John Temples) writes about automatic remapping: > The ESIX implementation catches errors while they're still "soft," > i.e., the error is recoverable. So remapping occurs with no data loss, > as long as the first time a sector has an error it isn't a hard error. I don't believe this is a common failure characteristic. Assuming that (1) you're running the drive in-spec, and (2) you've mapped out all the bad sectors determined by the drive manufacturer [N.B.: This is *NOT* the same as bad sectors found by running a r/w test], you shouldn't expect soft failures because you're not using any marginal sectors. A drive which is about to Bite the Big One may show a few soft errors before the disaster happens, but that's an omen that Something Bad Is About to Happen, so you want to know about it right away. For example, a tiny particle can get loose somehow. If it's just the right size to get under the head, it'll take a tiny ding out of the coating on a platter...and there's a good chance it'll be small enough to leave you with a soft error. However, you now have at least *two* tiny particles cruising around, possibly many more (the original and whatever got dug up). You can see how that one degenerates. It's only one hypothetical situation; the point is that if you start out using only the good sectors of a good disk and run in-spec, the sorts of things that can go wrong to produce soft errors are almost always (by that I mean something > 90%) precursors to a disastrous failure. If you run out-of-spec (e.g., non-RLL drives on an RLL controller), you're much more likely to see soft errors that stay soft. -- Dick Dunn rcd@ico.isc.com -or- ico!rcd Boulder, CO (303)449-2870 ...Simpler is better.