15 Aug, 2009, quixadhal wrote in the 22nd comment:
Interesting.
So, I've been doing some research into various drives to get for this little project, and came upon an issue related to modern drives and RAID systems that seems interesting.
Apparently, modern desktop drives have recovery logic built into their firmware which can put them into "deep recovery mode" on detection of a sector error, and that can last for up to 120 seconds. During that time, the drive is unresponsive, and thus will usually be dropped from most RAID systems with the appropriate failure indication.
Thing is, "RAID class" drives have this feature nerfed so that error recovery only lasts for 7 seconds, which means the error gets reported back quickly enough for the RAID system to not drop the drive. However, it also means the error remains an error, and isn't fixed (or shoved under the rug) by the drive itself.
Sooooo, there seems to be some contraversy over having this feature (Western Digital calls it TLER) enabled or disabled, and which types of RAID setups work in either case. RAID 5 systems seem to want this enabled (limited to 7 seconds), as the RAID system itself will handle the error (and perhaps try to rewrite the sector in queston?). RAID 0 systems, as well as desktops, want this disabled (full 120 second attempt) to avoid spurious errors cropping up.
The whole thing sounds like a pretty fishy way of saying "our drives suck compared to how we made them 20 years ago… but if you let us silently hide and fix errors, you'll think they work great!".
It's also worth noting that Western Digital refuses to accept warranty returns on desktop drives used in a RAID system, claiming you should have paid extra for the "RAID class" drive, even though the only real difference seems to be that setting (which you can toggle with a utility).
*sigh*
Makes me wish for our old VAX cluster. You wanted a drive? You called DEC. If you had any problems of any kind at all, DEC fixed it. Sure, you paid for it in service contracts… but I bet if you did the math, it cost less to pay those and have everything work and play nicely together, than companies spend nowadays with replaceable desktops, having IT staff "ghosting" drives all over the place, tech support calls to dozens of vendors, and the general hassle of trying to coordinate backups and where people actually put files they're working on.
So, I've been doing some research into various drives to get for this little project, and came upon an issue related to modern drives and RAID systems that seems interesting.
Apparently, modern desktop drives have recovery logic built into their firmware which can put them into "deep recovery mode" on detection of a sector error, and that can last for up to 120 seconds. During that time, the drive is unresponsive, and thus will usually be dropped from most RAID systems with the appropriate failure indication.
Thing is, "RAID class" drives have this feature nerfed so that error recovery only lasts for 7 seconds, which means the error gets reported back quickly enough for the RAID system to not drop the drive. However, it also means the error remains an error, and isn't fixed (or shoved under the rug) by the drive itself.
Sooooo, there seems to be some contraversy over having this feature (Western Digital calls it TLER) enabled or disabled, and which types of RAID setups work in either case. RAID 5 systems seem to want this enabled (limited to 7 seconds), as the RAID system itself will handle the error (and perhaps try to rewrite the sector in queston?). RAID 0 systems, as well as desktops, want this disabled (full 120 second attempt) to avoid spurious errors cropping up.
The whole thing sounds like a pretty fishy way of saying "our drives suck compared to how we made them 20 years ago… but if you let us silently hide and fix errors, you'll think they work great!".
It's also worth noting that Western Digital refuses to accept warranty returns on desktop drives used in a RAID system, claiming you should have paid extra for the "RAID class" drive, even though the only real difference seems to be that setting (which you can toggle with a utility).
*sigh*
Makes me wish for our old VAX cluster. You wanted a drive? You called DEC. If you had any problems of any kind at all, DEC fixed it. Sure, you paid for it in service contracts… but I bet if you did the math, it cost less to pay those and have everything work and play nicely together, than companies spend nowadays with replaceable desktops, having IT staff "ghosting" drives all over the place, tech support calls to dozens of vendors, and the general hassle of trying to coordinate backups and where people actually put files they're working on.
15 Aug, 2009, Igabod wrote in the 23rd comment:
It seems that the illusion of flawless functioning is just as good as actual flawless functioning… at least according to western digital. Though I suspect this theory has come into play in many facets of life. Sad really.
15 Aug, 2009, exeter wrote in the 24th comment:
It's not actually an illusion. The way these drives recover from and fix errors is via error-correcting codes, which is the same sort of technology NASA uses to ensure it receives good data from its space probes. (Incidentally, it's also the same technology your CD and DVD players use to be able to play a scratched disk.)
The mathematics behind ECCs is somewhat extensive, but what it all boils down to is that when you write data to the drive, the drive writes some additional data that gives enough additional information to recover your original data in the event your data is partially corrupted.
Also, re: RAID 5, I should say that I, personally, have lost far more data to stupid user errors (my own, that is) than I've ever lost to hardware failure. RAID does nothing to protect you from yourself. :P Rather than RAID, I'd personally go with a filesystem that has snapshot capabilities, like ZFS or btfs. Also, remember, RAID is not, in any way a substitute for backups. :-)
The mathematics behind ECCs is somewhat extensive, but what it all boils down to is that when you write data to the drive, the drive writes some additional data that gives enough additional information to recover your original data in the event your data is partially corrupted.
Also, re: RAID 5, I should say that I, personally, have lost far more data to stupid user errors (my own, that is) than I've ever lost to hardware failure. RAID does nothing to protect you from yourself. :P Rather than RAID, I'd personally go with a filesystem that has snapshot capabilities, like ZFS or btfs. Also, remember, RAID is not, in any way a substitute for backups. :-)
15 Aug, 2009, quixadhal wrote in the 25th comment:
RAID 1 or RAID 5 protect you from hardware failures… which is the only thing that's ever cost me data loss over the last 15 years or so. My own stupidity isn't data loss, that's brain cell loss – and no hardware will help with that until we get nanobots. :alien:
ZFS is a nice idea… but not for windows. I play games so I have a windows desktop. I also work with Microsoft development tools, as I'm hoping to code my way out of the poor house. If that fails, I'll still need those skills to get a job unless I happen to get really luck and find a place that wants a perl/postgresql/linux expert. Hasn't happened in the last 4 years, I'm not holding my breath.
I agree that you *SHOULD* do regular backups. I also *SHOULD* stop drinking sugary soda, start exercising before my arteries turn solid, and a bunch of other things that just never seem to happen with the kind of regularity that would make them worthwhile. So, since I recognize that I am lazy, and that backups happen on the rare occasion that I'm bored AND I think about it… a RAID system is better than nothing. At least if the magic smoke leaks out of a drive when the bearings melt, I have a second chance. ;)
ZFS is a nice idea… but not for windows. I play games so I have a windows desktop. I also work with Microsoft development tools, as I'm hoping to code my way out of the poor house. If that fails, I'll still need those skills to get a job unless I happen to get really luck and find a place that wants a perl/postgresql/linux expert. Hasn't happened in the last 4 years, I'm not holding my breath.
I agree that you *SHOULD* do regular backups. I also *SHOULD* stop drinking sugary soda, start exercising before my arteries turn solid, and a bunch of other things that just never seem to happen with the kind of regularity that would make them worthwhile. So, since I recognize that I am lazy, and that backups happen on the rare occasion that I'm bored AND I think about it… a RAID system is better than nothing. At least if the magic smoke leaks out of a drive when the bearings melt, I have a second chance. ;)
15 Aug, 2009, Guest wrote in the 26th comment:
Well I don't think very highly of Western Digital and their level of quality after having lost 6 of their drives back in 2007 within the span of the year. All warranty replaced of course, but fat lot of good that did for the data stored on them. This stuff Quix turned up seem to be a backhanded way for them to just admit they suck is all.
15 Aug, 2009, Runter wrote in the 27th comment:
At the company I worked at a few years ago we strictly used newegg and I never had a problem. Hard drives included.
15 Aug, 2009, Chris Bailey wrote in the 28th comment:
Newegg is an excellent place to shop. I kind of get screwed since they have a warehouse not far from me but won't allow pickup (I get the 9.75% TN sales tax though!). For future reference, if you guys want boxes inside your boxes, order retail parts instead of OEM. =)
15 Aug, 2009, David Haley wrote in the 29th comment:
Yo Chris, I heard you like boxes, so we put some packages in your packages so you can unpack while you unpack.
:blues:
:blues:
17 Aug, 2009, Hanz wrote in the 30th comment:
Newegg is an excellent place for any computer parts, I nearly bought an entire computer, part-for-part from them, and I haven't had any issues yet. :cool:
Me too, although I got very lucky in my case. A disk in my linux server died (14 year old WD 40g!), and I managed to find a couple of identical replacements in some old machines being thrown away at our local university (a friend of mine works there). So, I was able to swap circuit boards and get it working again.
Right after that, I installed a second of the "new" 40g drives and converted the live disk into a software RAID 1 array. That was educational. Now, I just have to remember that whenever the kernel gets updated, I have to install grub twice, once for the MBR of each drive so they're both bootable in case one dies in the future.
Right, but I think the vulnerability window is pretty small these days. This is a gaming machine, so I'm hoping the striping will help performance (I also keep non-game stuff on my machine, so RAID 0 is too risky for my taste), and I'm also hoping the OS will buffer writes enough so they either don't happen at all, or go through in a nice quick burst.
I would assume a hardware RAID controller would be smart enough to buffer a few cylinders worth of data so it can write out the checksums and data in one shot, rather than stupidly passing every block through as soon as it can.