lkml.org 
[lkml]   [2001]   [Nov]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    SubjectRe: Journaling pointless with today's hard disks?
    Date
    While I am not familiar with the IBM drives in particular, I am
    familar with this particular problem.

    The problem is that half of a sector gets new data, then when power is
    dropped, the old data+CRC/ECC is left on second part of that sector,
    and a subsequent read on the whole sector will detect the CRC/ECC
    mismatch, and determine the error burst is larger than what it can
    correct with retries, and ECC, and report it as a HARD ERROR. (03-1100
    in the SCSI World)

    Since the error is non-recoverable, the disk drive should not
    auto-reassign the sector, since it cannot succeed at moving good data
    to the newly assigned sector.

    This type of error does not require a low-level format. Just writing
    any data to the sector in error should give the sector a CRC/ECC field
    that matches the data in the sector, and you should not get hard
    errors when reading that sector anymore.

    This was more of a problem with older disk drives (8-Inch platters, or
    older), because the time required to finish any given sector was more
    than the amount of time the electronics would run reliably. All that
    could be guranteed on these older drives was that a power loss would
    not corrupt any adjacent data, ie write gate must be crow-bared
    inactive before the heads start retracting, emergency-style, to the
    landing zone.

    I believe that the time to complete a sector is so short on current
    drives, that they should be able to complete writing their current
    sector, but I do not believe that there are any drive manufacturers
    out there that gurrantee this. Thus, there is probably a window, on
    all disk drives out there, where a loss of power durring an active
    write will end up causing a hard error when that sector is
    subsequently read (I haven't looked though, and could be wrong).
    Writing to the sector with the error should clear the hard-error when
    that sector is read. A low-level format should not be required to fix
    this, and if it is, the drive is definitely broken in design.

    This is basic power-economics, and one of the reasons for UPS's

    Steve Brueggeman



    On 24 Nov 2001 14:03:11 +0100, you wrote:

    >In the German computer community, a statement from IBM[1] is
    >circulating which describes a rather peculiar behavior of certain IBM
    >IDE hard drivers (the DTLA series):
    >
    >When the drive is powered down during a write operation, the sector
    >which was being written has got an incorrect checksum stored on disk.
    >So far, so good---but if the sector is read later, the drive returns a
    >*permanent*, *hard* error, which can only be removed by a low-level
    >format (IBM provides a tool for it). The drive does not automatically
    >map out such sectors.
    >
    >IBM claims this isn't a firmware error, but thinks that this explains
    >the failures frequently observed with DTLA drivers (which might
    >reflect reality or not, I don't know, but that's not the point
    >anyway).
    >
    >Now my question: Obviously, journaling file systems do not work
    >correctly on drivers with such behavior. In contrast, a vital data
    >structure is frequently written to (the journal), so such file systems
    >*increase* the probability of complete failure (with a bad sector in
    >the journal, the file system is probably unusable; for non-journaling
    >file systems, only a part of the data becomes unavailable). Is the
    >DTLA hard disk behavior regarding aborted writes more common among
    >contemporary hard drives? Wouldn't this make journaling pretty
    >pointless?
    >
    >
    >1. http://www.cooling-solutions.de/dtla-faq (German)


    _________________________________________________________
    Do You Yahoo!?
    Get your free @yahoo.com address at http://mail.yahoo.com

    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2005-03-22 13:18    [W:0.030 / U:1.184 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site