> -----Original Message----- > From: Justin Piszcz [mailto:jpiszcz@xxxxxxxxxxxxxxx] > Sent: Thursday, December 18, 2008 10:43 AM > To: John Robinson > Cc: David Lethe; Linux RAID > Subject: Re: upgrade advice > > > > On Thu, 18 Dec 2008, John Robinson wrote: > > > On 17/12/2008 15:30, David Lethe wrote: > >> The duty cycle makes a difference now, but wasn't a design point > until > > [...] > >> OK, off my soapbox ... back to work writing disk diagnostic software > for > >> my OEM customers > > > > I am going to print that message, frame it and hang it on the wall. > > (Including the bit I elided with ...) > > > > I'll probably keep using desktop drives for domestic NAS, though. > I think it also depends on how often drives are used and what type of > workloads they are exposed to, for a domestic NAS for mainly sequential > file writes/reads, they will probably be OK-- I have a few SW RAID5's > on > desktop drives but I use them primarily as storage via rsync, once a > week > or month I am not constantly read/writing on them-- on a daily basis. > Is > that how you use your NAS as well? Or? > Surprisingly, the failure rates for consumer class ATA disks are counterintuitive. Breaking down into raw low,medium,high utilization ... this is what the study of 100,000 disks Revealed over 5 year period. (Rounded results, all in percentage) DISK AGE LOW MED HIGH (utilization) 3 months 4 2 10 6 mos 2 1 4 12 mos .5 1 2 2 years 2 2 2 4 years 3 4 4 5 years 1 1 5 So, if you pound a disk with I/O then it is 5x more likely to die in first 3 months then if it has light duty load. If a disk survives the first year, then load doesn't make much of a statistical difference ... until it approaches the 5 year mark. Note, these are for consumer class disks, that were better products back in 2000 when the study began. There are no long-term studies for real-world drive life for the (SATA) consumer vs. enterprise disk drives that are made today. The study involved 9 different disks, Seagate, Hitachi, WD, etc. that were typical of what you got with a personal computer from manufacturer, or you bought at a PC store. The test is as real-world as they come. As for drive temperature vs. failure rate, this will blow you away. The probability density curve is logarithmic in nature, so disks at 20 degrees C are 3X more likely to fail for any given load then disks at 26 degrees C. Once you hit 26 degrees then the curve really flattens out. Not much of a difference between 30-45 degrees. Sweet spot is 36-42 degrees. Colder disks are 6x more likely to fail then ones running around 40 deg. So don't throw money away on disk drive coolers/fans, unless temp without them is in the 45 deg C range! There is more correlation between failure rates of drives when both age and temperature are considered. The highlight that disks running colder than the sweet spot are 2-3x more likely to die in first 3 months then ones running near 40 deg. (AFR is 9% for coldest disks, 3% for warmest disks). >From 6 mos to 2 years, drive temp becomes less of a factor, with AFR of cold disks around 4%, sweet-spot disks 2%. Year 3, however is the killer, where everything changes... AFR of cold-warm disks approx 6 %, AFR of the disks from 35-40 deg is 11%. So aggregate walk-away - Keep disks in 36-42 degrees C for maximum life up to year 2, where they should be run cooler. - New disks are 5x more likely to die in first 3 months with high vs. low workload. - AFR and temperature become less important after the 3-month burn-in, but this all changes in year 3. David @ santools.com -- 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
