The Naked Scientists

The Naked Scientists Forum

Author Topic: How representative is the MTBF?  (Read 12170 times)

Offline CliffordK

  • Neilep Level Member
  • ******
  • Posts: 6321
  • Thanked: 3 times
  • Site Moderator
    • View Profile
How representative is the MTBF?
« on: 27/01/2012 22:30:51 »
Looking at new HDDs.

The Seagate ST33000650NS is rated as: 1,200,000 hours, or merely 136 years (for 50% failure rate).
It is also listed as an annual failure rate of 0.73.

The Hitachi Ultrastar, 7K3000 is rated at 2 Million Hours MTBF.
Or about 228 years of continuous duty (for 50% failure rate).
It doesn't seem to list an annual failure rate, but says it is 40% lower than the 1.2 Million hour drives, so it should be about 0.73*.6 = 0.44

I suppose that if I set up 2 drives in RAID 1 configuration...  that should do reasonably well to protect myself against HDD failures, although it would still be susceptible to anything that could take out 2 drives simultaneously (power surge, fire, water, etc).

So, should I expect these hard drives to last well into the next century without blowing any bearings, heads, magnetic media, or electronic gizmos?


 

Online syhprum

  • Neilep Level Member
  • ******
  • Posts: 3821
  • Thanked: 19 times
    • View Profile
Re: How representative is the MTBF?
« Reply #1 on: 27/01/2012 22:49:20 »
I wonder are these MTBF figures just for the drive spinning round doing nothing or continuiously chattering away.
I would like to see a figure for how many TB of data they can transfer before a faliure.
to the best of my knowlege satellites never use HDD's they did use tape recorders and now use SSD,s
« Last Edit: 28/01/2012 10:05:21 by syhprum »
 

Offline CliffordK

  • Neilep Level Member
  • ******
  • Posts: 6321
  • Thanked: 3 times
  • Site Moderator
    • View Profile
Re: How representative is the MTBF?
« Reply #2 on: 27/01/2012 23:22:11 »
One of the problems with tape media is that the media itself will degrade over time, at least in Earth's atmosphere.  So, a tape that is 10 or 20 years old, the magnetic stuff may just rub off (meaning lots of head cleaning, but also perhaps degradation of the data archives).  Perhaps Voyager & Pioneer don't have the same issues.  I must assume that the magnetic media has improved over time though.

SSD (Solid State Drives) are not without failures with some estimates that they are no better than hard drives.  However, many SSD crashes can be difficult to recover from (or perhaps need new recovery tools to be built).

As far as hard drives on a satellite, I believe that the drive heads require a pressurized atmosphere to hold them in place.  It isn't a big deal, but it would have to be part of the design of the satellite.  Without the pressurized atmosphere, one would likely get a miserable head crash almost immediately when using the drives in space, either that, or one could not control the head/platter distance.
 

Offline CliffordK

  • Neilep Level Member
  • ******
  • Posts: 6321
  • Thanked: 3 times
  • Site Moderator
    • View Profile
Re: How representative is the MTBF?
« Reply #3 on: 27/01/2012 23:33:55 »
As far as the MTBF, I believe that Hitachi does specify that it is for a "consumer" environment (24/7), and not a server environment. 

Certainly if one is doing a lot of memory swapping, one can exaggerate the load on the drive.

As far as 100+ years...  I would imagine it is all statistically derived based on the short-term failures of a small sample of drives.  The long-term success of the drive would depend on their bearings/bushings, and lubricants, and whether they would change over time. 

A company can certainly observe drive crashes since the 1970's or 1980's, and then fix the problems in old drives that show up.  But, my guess is that they actually plan for a 3-5 year commercial lifespan for the drives, and perhaps a 10 year consumer lifespan. 

So, the 100+ yr MTBF really wouldn't be indicative of problems that might crop up in as little as 20 years or so.
 

Offline Geezer

  • Neilep Level Member
  • ******
  • Posts: 8328
  • "Vive la résistance!"
    • View Profile
Re: How representative is the MTBF?
« Reply #4 on: 28/01/2012 04:31:14 »
They actually get a boat-load of drives and subject them to an accelerated life test. Temperature is usually the killer.
 

Offline CliffordK

  • Neilep Level Member
  • ******
  • Posts: 6321
  • Thanked: 3 times
  • Site Moderator
    • View Profile
Re: How representative is the MTBF?
« Reply #5 on: 28/01/2012 05:17:52 »
I think the company only has a 3 or 5 yr warranty.

But, if I got 2 drives configured as RAID-1...   Do you think I could convince them to keep replacing any drive that fails for the rest of my life....   As one would expect that if one had 2 drives, one should only get 50% failure after 228 years.

So, I would think it is false advertising if the drives don't last the rest of my life!!!
 

Online syhprum

  • Neilep Level Member
  • ******
  • Posts: 3821
  • Thanked: 19 times
    • View Profile
Re: How representative is the MTBF?
« Reply #6 on: 28/01/2012 10:04:38 »
I used to work with the old washing machine size 250MB drives the MTBF figures were a joke, I doubt if one ever ran for a year without a faliure of some sort.
« Last Edit: 28/01/2012 10:22:14 by syhprum »
 

Offline graham.d

  • Neilep Level Member
  • ******
  • Posts: 2208
    • View Profile
Re: How representative is the MTBF?
« Reply #7 on: 28/01/2012 20:01:22 »
You should look up the meaning of MTBF. It says nothing about total operating life of the units. It means,  for say an MTBF of 2,000,000 hours (about 228 years), that for every 228 units operating there will be an average of 1 failure per year. The operating life of the unit may be just 3 years but that does not change the MTBF value.
 

Offline Geezer

  • Neilep Level Member
  • ******
  • Posts: 8328
  • "Vive la résistance!"
    • View Profile
Re: How representative is the MTBF?
« Reply #8 on: 28/01/2012 23:55:14 »
I know what Syhprum is referring to. We used "front loader" disks (I think they had a whopping 5 megabyte capacity) and they would drop like flies. On one occasion I was leaning against one when it decided to crash its heads. It sounded exactly like a gigantic circular saw chewing up a tree!

MTBF is a very interesting subject (at least reliability engineers seem to think it is). There's a fair bit of science that goes into determining MTBF, but basically what you do is get a large number of samples and beat the **** out of them under arduous conditions.

Thereafter, the design engineers and the reliability engineers get into a major pissing contest about the validity of the tests while the executives hover around as their bonuses evaporate.
 

Offline SeanB

  • Neilep Level Member
  • ******
  • Posts: 1118
  • Thanked: 3 times
    • View Profile
Re: How representative is the MTBF?
« Reply #9 on: 29/01/2012 06:20:05 »
Just to add to this, now the warranty on drives is 1 year, down from the 5 years it was before. Most of the time the electronics fail before the drive, indicating a poor design or cost cutting on the manufacturers side. I always place extra fans directed at the drive, so as to keep the electronics cool.
 

Online syhprum

  • Neilep Level Member
  • ******
  • Posts: 3821
  • Thanked: 19 times
    • View Profile
Re: How representative is the MTBF?
« Reply #10 on: 11/02/2012 23:09:37 »
I have seen some life figures publshed on a Hitaichi SSD 512GB device they claim it can handle A thruput of 19.2 TB per day for five years, pretty impressive and the right way to specify drive life.
 

Offline CliffordK

  • Neilep Level Member
  • ******
  • Posts: 6321
  • Thanked: 3 times
  • Site Moderator
    • View Profile
Re: How representative is the MTBF?
« Reply #11 on: 11/02/2012 23:51:24 »
5 years?  What happens in 5 years?
That is a lot of data being recorded, unless the SSD is being used as a memory accelerator ramdrive with permanent resident data.  A true Ramdrive with a backup might be a better choice for use as an accelerator.

One of the past issues with SSD's was "Write Endurance", but apparently the write endurance issue has been improved by maintaining a log of the number of write cycles per block, dynamically distributing writes, and better hardware.

However, reviews seem to indicate an unacceptably high "return rate" or "failure rate" among SSDs, so 5 year failure might not be unexpected.
 

Offline SeanB

  • Neilep Level Member
  • ******
  • Posts: 1118
  • Thanked: 3 times
    • View Profile
Re: How representative is the MTBF?
« Reply #12 on: 12/02/2012 11:23:43 »
SSD reliabilty is improving with wear levelling, but the issue is that areas will fail, hopefully going unreliable before dead so that you can get the data off. The problem is the areas that get used the most like the file descriptors, where a few hundred writes can be made for every file, as blocks are written to the disk elsewhere. As well the relocated sectors need an area that contains the relocation table, and this cannot be relocated as it is read at power on and must be a specific location. This table is invisible to the outside.

A while ago I decided to kill a 2G USB memory stick by installing a Linux filesystem on it, complete with a swap partition. I left it running idle, with only the normal logging system running, no screensaver on and nothing open. Lasted 3 days before it exceeded the write cycles of the area used for the swap partition and the logs and did a kernel panic. On testing the drive a large number of areas had failed to write.
 

Offline graham.d

  • Neilep Level Member
  • ******
  • Posts: 2208
    • View Profile
Re: How representative is the MTBF?
« Reply #13 on: 12/02/2012 11:58:06 »
A lot of memory sticks do not have good wear levelling incorporated into them. Wear levelling is (necessarily) done at a low level within the flash memory controller chip and, although you could get away with not having it on NOR based flash, it is essential for any realistic usage of NAND flash. Very essential for Flash based disk drives, even with NOR flash, or you get the problem as you get unrecoverable failures very quickly.
 

The Naked Scientists Forum

Re: How representative is the MTBF?
« Reply #13 on: 12/02/2012 11:58:06 »

 

SMF 2.0.10 | SMF © 2015, Simple Machines
SMFAds for Free Forums
 
Login
Login with username, password and session length