[geeks] bah.

Phil Stracchino alaric at metrocast.net
Fri Mar 13 11:54:50 CDT 2009


Jonathan C. Patschke wrote:
> On Thu, 12 Mar 2009, Phil Stracchino wrote:
> 
>>> The geom stuff... it has burned me a lot.  The geom raid software will
>>> happily destroy all your data with no warning either from the programs
>>> or the man pages, and they frequently locked up FreeBSD 6 and 7.  I
>>> have not tried FreeBSD 7.1 yet so things may have improved in that
>>> release.
>> Huh.  I used the recommended geom method for boot disk mirroring ...
>> wonder if that was part of the problem?  And considering it was
>> mirroring SATA disks, that sounds like a possible double whammy.
> 
> Knocking on wood, I have one system that has a pair of geom-mirrored
> disks, has seen constant (sometimes very heavy) disk and network activity
> since it was turned on 696 days ago, and has absolutely never gone
> unavailable.

Actually, my "most probable cause" theory about the stability issues is
that the 3Ware board is shaky.  System stable under light load, start
hitting the array hard [say, start copying a third of a terabyte of data
to it] and the machine falls over ... I'd suspect the disk controller,
wouldn't you?

Oh, for a nice pair of LSIs....

In the meantime, once I can get the BIOS flashed to v1.08 I'll be trying
the FreeBSD route again and see if it'll boot amd64 then.


> Let me qualify my praise of geom with a stern damning stemming from its
> needlessly spartan documentation and overall lack of discoverability.  If
> you don't know that it's there (and up until FreeBSD 7 or so, you wouldn't
> know unless you asked someone who knew), you won't stumble across it.
> And, even if you do stumble across it, the handbook and man pages don't go
> into a lot of depth.
> 
> For example, the gjournal manpage nowhere tells you that if you have
> enough I/O activity to overrun the journal, your box will panic.  That's
> the sort of things folks might want to know, just perhaps.


Huh.  Yeah, that'd be an important thing to know ....  but that
shouldn't be relevant to heavy I/O activity on the storage array managed
by ZFS, not on the boot pair managed by geom, should it?


-- 
  Phil Stracchino, CDK#2     DoD#299792458     ICBM: 43.5607, -71.355
  alaric at caerllewys.net   alaric at metrocast.net   phil at co.ordinate.org
         Renaissance Man, Unix ronin, Perl hacker, Free Stater
                 It's not the years, it's the mileage.



More information about the geeks mailing list