DIY HDD platter swap?

[Mechanical Menace]

I sure hope not ..

Maybe I've just had a bad experience with them. They were identical drives bought at the same time so likely, and I wasn't sold them as Toshiba either. They were Toshiba drives in some German company I've never heard of boxes, which I was sold as Hitachi's but couldn't get my money back.

this is Toshiba DT01ACA300 aka rebadged Hitachi

Explains the circumstances of me receiving them. Mine were the DT01ACA100 if that makes much of a difference.

EDIT:Nope it was the 300, was reading the wrong part of the label lol. Says it was from April 13.

[wraper]

There is lots of misinformation on the net regarding reallocated sector counts.  It really is normal for large TB disks to have some reallocated sectors, in my opinion it's inevitable.  I think I might have dropped the ball when it started growing over the years. There was a huge growth at the beginning when it was first installed, and then it was stable for a few years, but recently would add one or two more sectors every month or two.

As much as I have read, growing reallocated sectors from the very beginning is quiet specific for Seagate, which is one more reason to not buy them.

[PlainName]

bought 2 new Toshiba 2TB drives to replace

A bad move, IMO. When I set up a mirror I always use different manufacturers. Doesn't matter if one of them has a bad rep - you're planning for a fail and if it fails way before the other drive that's good! Just so long as the fail cycles don't get in sync...

[madires]

Please use server/RAID disks for servers and not desktop disks! Desktop disks are not designed to run 24x7 with a typical server usage profile.

[codeboy2k]

... Just so long as the fail cycles don't get in sync...

... as two of my 3 Seagates did ! (all bought at the same time)

I have already bought 2 Hitachi drives DT01ACA200.  I think I will add to this 2 more WD drives and split them across a RAID10, with 1 Hitachi drive mirroring a WD drive, then striping across 2 mirror sets like that.  If both Hitachis fail at the same time, I still have 2 WD in the stripe.

                            POOL
              =================
===========              ===========
[ WD ] [ Hit ]                [ WD ] [ Hit ]

Please use server/RAID disks for servers and not desktop disks! Desktop disks are not designed to run 24x7 with a typical server usage profile.

It's my home server.  It's lightly used, and runs 24x7 just like my desktop.  By no means is it seeing a typical corporate server usage profile, and desktop drives are fine.  I just need to properly plan for failure.

[sync]

That's clear now

There is lots of misinformation on the net regarding reallocated sector counts.  It really is normal for large TB disks to have some reallocated sectors, in my opinion it's inevitable.

A new disk has 0 reallocated sectors reported via SMART. The factory defect sectors are not shown. A few grown reallocated sectors are not a problem. It can happen. But they should not constantly growing. I would not trust a disk which has more than 10-20 reallocated sectors (after a few years).

This is a 4 years old 1TB hard drive (WD10EARS) from my XBMC PC:
  1 Raw_Read_Error_Rate     0x002f   200   200   051    Pre-fail  Always       -       0
  3 Spin_Up_Time            0x0027   189   183   021    Pre-fail  Always       -       6533
  4 Start_Stop_Count        0x0032   098   098   000    Old_age   Always       -       2582
  5 Reallocated_Sector_Ct   0x0033   199   199   140    Pre-fail  Always       -       1
  7 Seek_Error_Rate         0x002e   200   200   000    Old_age   Always       -       0
  9 Power_On_Hours          0x0032   094   094   000    Old_age   Always       -       4433
 10 Spin_Retry_Count        0x0032   100   100   000    Old_age   Always       -       0
 11 Calibration_Retry_Count 0x0032   100   100   000    Old_age   Always       -       0
 12 Power_Cycle_Count       0x0032   098   098   000    Old_age   Always       -       2581
192 Power-Off_Retract_Count 0x0032   200   200   000    Old_age   Always       -       61
193 Load_Cycle_Count        0x0032   075   075   000    Old_age   Always       -       375033
194 Temperature_Celsius     0x0022   114   098   000    Old_age   Always       -       33
196 Reallocated_Event_Count 0x0032   199   199   000    Old_age   Always       -       1
197 Current_Pending_Sector  0x0032   200   200   000    Old_age   Always       -       14
198 Offline_Uncorrectable   0x0030   200   200   000    Old_age   Offline      -       7
199 UDMA_CRC_Error_Count    0x0032   200   200   000    Old_age   Always       -       0
200 Multi_Zone_Error_Rate   0x0008   200   200   000    Old_age   Offline      -       21

Only one reallocated sector but 7 Offline Uncorrectable ones :-(

btw: I try to keep the disks in servers/RAID systems under 30°C. Which is usually not a problem with well designed rack mount servers and cooled server rooms. Not easy at home of course.

And as madires said, use server disk for anything which runs 24/7 or the data is important. Enterprise 7200rpm SATA disks are affordable.

[codeboy2k]

That's clear now

There is lots of misinformation on the net regarding reallocated sector counts.  It really is normal for large TB disks to have some reallocated sectors, in my opinion it's inevitable.

A new disk has 0 reallocated sectors reported via SMART. The factory defect sectors are not shown. A few grown reallocated sectors are not a problem. It can happen. But they should not constantly growing. I would not trust a disk which has more than 10-20 reallocated sectors (after a few years).

Really so low?  If that's the case, then I'll need to change my thinking. I'm a techie, I have never been an enterprise server admin, so I don't have any first-hand experience with disk drive statistics or when to know when it's going bad.  You say 20 sectors on a 2TB disk that reports about 3907029168 sectors (512B/sector), but is only 0.5 ppm or 500ppb. It's really, really, small to be throwing the disk out, isn't it ? 

And as madires said, use server disk for anything which runs 24/7 or the data is important. Enterprise 7200rpm SATA disks are affordable.

I'm going to rebuild this server with a mix of the 2 x  2TB Hitachi's that I already bought and 2 more  2TB WD red's these are $119 and I agree are affordable.
For future backup, I've decided to get a used LTO3 drive from eBay. LTO3 tapes are cheaper and more space efficient than buying more disk, and if I grow the server I just need to add more tape.  Plus tape allows me to get a real monthly/weekly/daily strategy going without a lot of cost.

I'll try to rebuild the lost RAID offline after I get this server back up with all new drives.

[wraper]

Really so low?  If that's the case, then I'll need to change my thinking. I'm a techie, I have never been an enterprise server admin, so I don't have any first-hand experience with disk drive statistics or when to know when it's going bad.  You say 20 sectors on a 2TB disk that reports about 3907029168 sectors (512B/sector), but is only 0.5 ppm or 500ppb. It's really, really, small to be throwing the disk out, isn't it ?

How much worth is in the HDD which corrupts data may I ask? Reallocated sector = sector was remapped because data couldn't be read from it = your data were corrupted. Now let's imagine, your HDD is full of HD video, each file like a 40-50 GB. Now take your 0.5ppm of corruption figure and suddenly it appears that you don't have even a single file unaffected and all of them were corrupted.

[codeboy2k]

3TB toshiba 4800 hours.

That Toshiba is half way through it's life span if my experiences are at all common.

this is Toshiba DT01ACA300 aka rebadged Hitachi HDS723030BLE640 which proved to be quiet reliable.

I did some googling, and it would seem that it's also the IBM Deskstar 7K3000.  Reviewers say it's a 5-platter drive, which would account for it's reliability. Most 3TB drives are 3 or 2 platters these days, which means the bits are denser and harder to read, so it can fail more easily.  That's going to be my opinion going forward from now on.

I was originally going to do a RAID10 with a mix of 2TB Hitachi and 2TB WD red's (4TB total) but I think I'll change my mind now, and simply do a 2 disk mirror of 1 2TB Toshiba/Hitachi DT01ACA200 and 1 2TB WD red.  (for 2 TB total).  It's slightly cheaper per TB ($86 per TB vs $100 / TB), uses less power than 4 drives, and good enough for me right now as I don't have the data anymore to need more than 2 TB storage. 

And if I'm thinking of LTO3 tape backup, do I even need to mirror it anymore? Why not just stripe it for 4 TB total and rely on (tested!) backups then. It's my home server, I really don't NEED the reliability of RAID, I realize that now.  What I need, going forward, is good backups.

[codeboy2k]

How much worth is in the HDD which corrupts data may I ask? Reallocated sector = sector was remapped because data couldn't be read from it = your data were corrupted. Now let's imagine, your HDD is full of HD video, each file like a 40-50 GB. Now take your 0.5ppm of corruption figure and suddenly it appears that you don't have even a single file unaffected and all of them were corrupted.

It's just my home data.  No business data, and nothing that I can't do without right now, which is why I'm not rushing the disk to an $800 data recovery service.

What I will do is rebuild the server, get a good backup system in place, and go forward.  The lost data on my existing RAID I'll keep those disks separate and offline, and try to recover any data just as a learning exercise.

[wraper]

Home or business data. What is the value of the medium if it does not serve it's main purpose - store the data intact. Imagine some firmware with 0.5 ppm of data corrupted. like a single byte corrupted in 2 MB of flash ROM. Your device becomes useless junk.
http://www.trademe.co.nz/Browse/Listing.aspx?id=925187196

Good for paperweight

[TerraHertz]

I've completely given up on RAID systems.
* They add complexity in setup and management,
* Add an extra critical failure point - the RAID system hardware itself. If that fails the drives are typically unreadable via other means, and trying to replace the failed board or whatever can be hard &/or expensive.
* You typically can't just pull a drive out and read or copy it using standard hardware.

The simple rule I follow now is, can I read the drive in a cheap external USB HDD dock? Can I do complete file system duplications using treecopy utilities that copy all files without corrupting attributes like file dates, and will run to completion without throwing millions of errors and stopping without being restartable?

I won't use any system that doesn't allow that.
I keep all 'work' drives in removable trays, and never ever mix system/utility installs with work spaces.
With that, I can manage work backups as appropriate, using any old excess/found/salvaged drives or USB sticks (free), and all of the backup devices are readable as standard file systems on any PC. Such backups can take a long time, but I can run them on any spare PC without tying up my main work system.
Also when possible I format backup drives as FAT32, since as a last resort I can manually examine and patch FAT32, but NTFS is completely beyond such measures. For me, anyway. It sucks that now almost always filesets are too big for FAT32 drives.

Incidentally, many years ago, maybe around 2004, I tried a HD PCB & platter swap. The drive contents was fairly important. I'd decided a backup was way overdue, and was spending a few hours doing folder tree structure tidying, before backing up. Ha ha, big mistake inviting Mr Murphy to lunch. During that process the drive suddenly became a non-drive, ie not recognized by the BIOS at all, suggesting PCB failure. I had an identical working blank drive. Not sure if removing the PCB might break hermetic seal of the interior, I improvised a 'clean area' by using a large clear plastic bag, new from a stack and never 'fluffed up', so hopefully the interior was dust free. Cleaned the two drive exteriors with compressed air, slipped them and a screwdriver into the bag, taped surgical gloves onto the bag opening, and swapped the good PCB onto the bad drive. Result: now it was recognized, but couldn't read any data. I then tried a platter swap, so the single precious platter was in the body with known good PCB, heads and head-amp. That wasn't readable either.

On inquiring, I was told that (even then) the boards had head tracking calibration data in flash, and board or platter swaps never worked anymore. Never did recover the lost files.

[Halcyon]

I've never seen anyone successfully recover a HDD by swapping either the PCB or the platters with a good drive.

We do just this at work when other methods fail. We use a "donor drive" to swap the PCB out of. But it doesn't always work (even minor differences in drive firmware is sometimes enough for it to fail). If we get really desperate, the drive gets sent off to another Government lab with lots of cool and expensive toys. :-)

Having lost data myself without a proper backup routine in place (not due to drive failure, but I locked myself out of the encrypted drive as the password was stored in a KeePass file inside the encrypted volume, do'h! Catch 22). Ever since, I back everything up to disk and LTO tape.

[sync]

Really so low?  If that's the case, then I'll need to change my thinking. I'm a techie, I have never been an enterprise server admin, so I don't have any first-hand experience with disk drive statistics or when to know when it's going bad.  You say 20 sectors on a 2TB disk that reports about 3907029168 sectors (512B/sector), but is only 0.5 ppm or 500ppb. It's really, really, small to be throwing the disk out, isn't it ? 

Yes, really so low. Also this is for a desktop PC like case. On an enterprise server I didn't care. It's the RAID controller's job. Don't think in relative terms like ppm here. Every reallocated sector is a little fault. Would you trust a device which is constantly getting new faults?

For future backup, I've decided to get a used LTO3 drive from eBay. LTO3 tapes are cheaper and more space efficient than buying more disk, and if I grow the server I just need to add more tape.  Plus tape allows me to get a real monthly/weekly/daily strategy going without a lot of cost.

LTO is a nice technology. But tapes have also a few drawbacks. First I'm skeptical about used tape drives from ebay. Tape drives will wear out. They are delicate precision mechanically devices. They should be used in a dust free environment. They want a minimum sustained data rate or it stop, reverse a bit and start again repeatably (shoe-shining). This will wear out the drives and tapes fast. Also this reduces the usable capacity. Depending on the LTO-3 drive at leased 60 or 70MB/s sustained data rate is need. Better double for compression.

The capacity of a LTO-3 tape is 400GB, compressed up to 800GB. You are taking about a few TB of storage. If you want to backup that you need a lot of tapes. Which means swapping tapes over hours for a full backup.

I'll try to rebuild the lost RAID offline after I get this server back up with all new drives.

Good luck with that.

[senso]

I've never seen anyone successfully recover a HDD by swapping either the PCB or the platters with a good drive.
It probably depends on the brand, some might work, but i think most drives have their low level configuration split between flash memory on the PCB and data on the platters.
So swapping things around just corrupts stuff.  I can see it working if the motor is dead though, (swapping both the pcb and platters).

Done that, successfully, just remember to also swap the SOIC-8 EEPROM and its all good(when its a pcb related damage, in the case, it was a grilled motor driver).

[Psi]

Here's my oldest drive that's still in use, happens to be a seagate (before they got shit).
It's in my server and has been running 24/7 for 87998 hours... 10 years!!!
It's seen entire generations of other seagate models (1.5 and 3TB) get designed, manufactured, and then die.
And it still doesn't have any reallocations or even any pending reallocations.

=== START OF INFORMATION SECTION ===
Model Family:     Seagate Barracuda 7200.8
Device Model:     ST3200826A
Serial Number:    3ND0HW6Y
Firmware Version: 3.02
User Capacity:    200,049,647,616 bytes [200 GB]

Vendor Specific SMART Attributes with Thresholds:
ID# ATTRIBUTE_NAME          FLAG     VALUE WORST THRESH TYPE      UPDATED  WHEN_FAILED RAW_VALUE
  1 Raw_Read_Error_Rate     0x000f   052   048   006    Pre-fail  Always       -       78530478
  3 Spin_Up_Time            0x0003   098   098   000    Pre-fail  Always       -       0
  4 Start_Stop_Count        0x0032   100   100   020    Old_age   Always       -       421
  5 Reallocated_Sector_Ct   0x0033   100   100   036    Pre-fail  Always       -       0
  7 Seek_Error_Rate         0x000f   064   060   030    Pre-fail  Always       -       2810645
  9 Power_On_Hours          0x0032   001   001   000    Old_age   Always       -       87998
 10 Spin_Retry_Count        0x0013   100   100   097    Pre-fail  Always       -       0
 12 Power_Cycle_Count       0x0032   100   100   020    Old_age   Always       -       511
194 Temperature_Celsius     0x0022   029   045   000    Old_age   Always       -       29 (0 14 0 0 0)
195 Hardware_ECC_Recovered  0x001a   052   047   000    Old_age   Always       -       78530478
197 Current_Pending_Sector  0x0012   100   100   000    Old_age   Always       -       0
198 Offline_Uncorrectable   0x0010   100   100   000    Old_age   Offline      -       0
199 UDMA_CRC_Error_Count    0x003e   200   200   000    Old_age   Always       -       3
200 Multi_Zone_Error_Rate   0x0000   100   253   000    Old_age   Offline      -       0
202 Data_Address_Mark_Errs  0x0032   100   253   000    Old_age   Always       -       0

[SeanB]

I did it with a 20G Seagate, luckily I had bought a few of the same model drive at the time ( same size Seagate, and the donor was a very close firmware revision drive), though the temporary donor machine was taken off of a very unhappy data capture clerk for a 2 day period, as her machine was the closest match. Got the 10G of data I needed and the OS cloned to another ( slightly newer but still used) drive, before I swapped controllers again to return the machine.  cloned the drive again so I would have a spare, as the software is no longer obtainable, but is happy to run with a dongle only. Finally retired the Win95 machine last year, as the PBX it ran was replaced with a newer system. Still have it, and it still works, though I would not want to use IE4 again on the internet, and it is only a 90MHz Pentium processor. I kept it running using the pile of old scrapped slowpoke machines over the years, upgrading RAM to 512M as I got the modules as opposed to the 64M it originally had.

Still have a pile of 32M, 64M, 128M SIMM's around in a baggie, I used a few in printer upgrades, where they fitted older HP printers. Some went to being able to print a full page graphic from the old RAM printing a quarter page at a time.

[SeanB]

Here's my oldest drive that's still in use, happens to be a seagate (before they got shit).
It's in my server and has been running 24/7 for 87998 hours... 10 years!!!
It's seen entire generations of other seagate models (1.5 and 3TB) get designed, manufactured, and then die.
And it still doesn't have any reallocations or even any pending reallocations.

Same with the 40G drive in an old laptop I use as a desktop spare machine at work. I run it as a media player, and use it to check internal network only so I can simply use the web interfaces of printers to save walking there to check the "It has a red light and won't print my job WAAAH and I can't be bothered to read the FSCKING display on the front to see what the error is", so I can take a cartridge where needed, or put paper in the machine. Yes there are people who think that they work by making paper and toner out of thin air, and complain they can't print when the power is out.

[amyk]

  4 Start_Stop_Count        0x0032   100   100   020    Old_age   Always       -       421

It's been running on average for over 200h between each spinup/spindown. That contributes to longevity, as start/stop cycles cause the most wear.

I also follow a "zero tolerance" policy on reallocated sectors - if any new ones develop after the drive has been in use, I replace the drive since more are coming.

[BradC]

Spinrite does a think called dynostat recovery

Don't get me started on Spinrite - I'm convinced it doesn't work.  Steve Gibson is a really smart guy, and I bet back when he came up with the software (1980's) if worked 50% of the time when drives had the aerial density of a the hindenburg - but nope - I've tried it on about 25 drives I'd say - at least 20.  I've never recovered a drive with it - ever - not even a single file.

Finally a topic I can contribute to. Back in about 2003 as Mr Gibson was finalising SR6 I spent a long time running it in an instrumented VM to actually watch its interaction with the ATA controller.

Let's get a couple of things straight. Spinrite has not performed magic for about 20 years. It does not and can't "disable SMART" and it does not interact or interface with the drives firmware (or any other bullshit people say it does). What it does do (and I'm mystified about why nobody else is doing this) is invoke "dynastat" recovery.

This does work for one specific failure. You have a PC that dies (GPF, bluescreen, hang, whatever) when it hits a bad sector on the disk. Spinrite reads the disk end to end and when it hits the dud it invokes "Dynastat". This reads the dud sector using the ATA read_long (and sometimes some other low level commands to the same effect) repetitively. This command returns the raw sector contents in whatever form they come off the disk having failed ECC recovery. If in the process the disk gets a good read, the disk itself will re-write that sector. Problem solved. In most cases it won't, so Spinrite builds a simple statistical model of each bit in the sector. It does this X times (where X can be defined on the command line) and at the end of X times it writes the statistical model back to the sector on the disk. In general this writes a "best guess" of the sector contents back and in most cases it solves the problem.

On a failing disk, Spinrite will just run it into the ground and trash it completely, but for mum's PC which is just failing to boot it's a quick fix that makes the problem go away.

So it does work, for certain use cases based on old technology and smaller capacity drives. It is useless for "preventative maintenance" or other mystic magic Steves followers promote it for, but hey people buy snake oil and put it in their cars, or monster cable. At least this *does* do something for specific cases.

[max666]

Please use server/RAID disks for servers and not desktop disks! Desktop disks are not designed to run 24x7 with a typical server usage profile.

Not only that, but drives in a server (or any multi-bay enclosure) also have to endure higher vibrations from all the closely mounted neighbouring drives, which are happily humming and clicking away.
However I myself don't follow that advice, because I'm too cheap. But I also had surprisingly little issues with my countless desktop drives running in RAID5 24x7.

[wraper]

Found on Russian forum that in HDDs of these series a seal under connector becomes leaky.

How to bring a second life to it 

[Dave Turner]

I was part of the team running a small/medium sized company's network  from '98 to '09. The network and server farm expanded considerably during this period. The one thing we tried to do was mix batches of disks for the raids, mirrors and SAN racks. I don't recall the source of the information that caused me to do this, however we never lost information due to the relatively few failed server drives we suffered.

If you think about it, it makes 'gut' sense to mix drive sources provided they have identical properties, and whilst I can't work the probability statistics out I'm sure there are those that can.

There is a reason for not being on the 'bleeding edge' just as there is for not using a 'dot zero' product when one's business depends on very high reliability.

Dave

[edavid]

Please use server/RAID disks for servers and not desktop disks! Desktop disks are not designed to run 24x7 with a typical server usage profile.

Is there any evidence that server/RAID drives have lower failure rates?

[sync]

Please use server/RAID disks for servers and not desktop disks! Desktop disks are not designed to run 24x7 with a typical server usage profile.

Is there any evidence that server/RAID drives have lower failure rates?

Does over 15 years experience count?