DragonFly users List (threaded) for 2009-06
DragonFly BSD
DragonFly users List (threaded) for 2009-06
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: Dragonfly used in production


From: Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx>
Date: Mon, 22 Jun 2009 11:22:14 -0700 (PDT)

:> Hammer does not have have long fsck periods; that was one of the design
:> goals.
:> 
:> There's no software RAID as part of Hammer right now; you could use CCD or
:> maybe vinum?  This isn't directly helpful, but I usually favor hardware
:> RAID when possible.
:
:	If you are using SATA (or IDE) drives you should be able to use
:nataraid (see nataraid and natacontrol man pages).
:
:-- 

    I'm not really happy with any of our current solutions, though vinum
    comes closest to fitting the bill.  The fake-raid stuff is too
    vendor-dependant..  I'd rather just have a small boot partition
    on every disk in the system or, if it were really that important, one
    could simply boot off of a solid state drive which has the same MTBF
    as the computer's motherboard.  Booting off of a SSD is far more
    reliable then nataraid with 2 or 3 normal hard drives.

    From my perspective, then, this means that kernel-supported software
    RAID is the way to go.  e.g. vinum (if it could be made reliable enough),
    or we could work up our own solution.  The mechanics of a software
    RAID system are not all that complex in actual fact.  In fact,
    soft-raid mirroring would work quite nicely with HAMMER because HAMMER
    already deals with any inconsistencies which might occur due to a
    system crash (e.g. due to the soft-raid not being able to complete
    an I/O operations across all the disks making up the operation).
    On system recovery HAMMER will run its undo and thus re-write any
    inconsistent sectors.  A soft-raid system will thus see the re-writes
    and thus be able to recover the consistency for those particular sectors.

    This means the soft-raid system would only really have to deal with
    inconsistencies which build up due to an actual disk failure.  It
    would NOT have to deal with transactional inconsistencies due to a
    system crash.  Anyone who knows how RAID works should know that not
    having to deal with transactional inconsistencies reduces the complexity
    of the RAID implementation by an order of magnitude.

    I think this is a more appropriate way to separate the RAID
    functionality from the filesystem functionality.  The FS has just
    enough logic in it to deal with inconsistencies due to system
    crashes and the RAID system only has to deal with inconsistencies
    due to actual disk failures.  Win. Win.

					-Matt
					Matthew Dillon 
					<dillon@backplane.com>



[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]