> Can you explain how the disks have a MTBF of 1,000,000 hours? But fail more > often than that? Maybe I just don't understand some aspect of MTBF. simple: the MTBF applies to very large sets of disks. if you had millions of disks, you'd expect to average mtbf/ndisks between failures. with statistically trivial sample sizes (10 disks), you can't really say much. of course, a proper model of the failure rate would have a lot more than 1 parameter... for instance, my organization will be buying about .5 PB of storage soon. here are some options: disk n mtbf hours $/disk $K total 250GB SATA 1920 1e6 500 399 766 600GB SATA 800 1e6 1250 600? 480 73GB SCSI/FC 6575 1.3e6 198 389 2558 146GB SCSI/FC 3288 1.3e6 395 600 1973 300GB SCSI/FC 1600 1.3e6 813 1200 1920 these mtbf's are basically made up, since disk vendors aren't really very helpful in publishing their true reliability distributions. these disk counts are starting to be big enough to give some meaning to the hours=mtbf/n calculation - I'd WAG that "hours" is within a factor of two. (I looked at only three lines of SCSI disks to get 1.3e6 - two quoted 1.2 and the newer was 1.4.) vendors seem to be switching to quoting "annualized failure rates", which are probably easier to understand - 1.2e6 MTBF or 0.73% AFR, for instance. the latter makes it more clear that we're talking about gambling ;) but the message is clear: for a fixed, large capacity, your main concern should be bigger disks. since our money is also fixed, you can see that SCSI/FC prices are a big problem (these are real list prices from a tier-1 vendor who marks up their SATA by an embarassing amount...) further, there's absolutely no chance we could ever keep .5 PB of disks busy at 100% duty cycle, so that's not a reason to buy SCSI/FC either... regards, mark hahn. - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
