Re: WAPBL

[buhrow%lothlorien.nfbcal.org@localhost (Brian Buhrow)]

        hello.  My understanding is that Greg has done some of this work
already.  He helped me solve a problem which was  causing large disks to
fail to reconstruct after a disk failure.  The failure was due to an
over-use of kernel memory to keep track of all the parity regions.  So,
while I don't know much about this topic, I'm hopeful that we're somewhat
along the path to this happy day.  Reducing the time it takes to clean
parity on a raid set  would be a  beautiful thing.
-Brian
On Aug 1, 12:17am, Thor Lancelot Simon wrote:
} Subject: Re: WAPBL
} On Thu, Jul 31, 2008 at 09:48:00PM +0100, Matthias Scheler wrote:
} > 
} > Any brilliant ideas how to teach RAIDframe (ETA for parity rewrite
} > is 1:28 hours) to perform only limited parity rewrites after a crash?
} 
} I think the parity logging code could be enhanced to do this (in addition
} to providing the significant performance benefit it should already provide).
} It divides the volume into "parity regions" already, so it should be possible
} to add a clean bit per parity region, set after each parity log transaction
} completes.
} 
} Greg says the parity log code has some problems with large disks -- it
} doesn't use large enough regions (they seem to be fixed size), so it
} wants e.g. 1GB of kernel memory per 1TB of managed disk.  But that could
} presumably be addressed.
} 
} I am *not* volunteering to do this.
} 
} -- 
} Thor Lancelot Simon                                      
tls%rek.tjls.com@localhost
}  "My guess is that the minimal training typically provided would only
}  have given the party in question multiple new and elaborate ways to do
}  something incomprehensibly stupid and dangerous."    -Rich Goldstone
>-- End of excerpt from Thor Lancelot Simon