kern/92518: Intense disk activity (large FS newfs and/or many
small files copy): DMA completion error with RocketRAID-1820A
and FastTRAK-S150
phillip at it.jcu.edu.au
phillip at it.jcu.edu.au
Thu Feb 16 04:40:22 PST 2006
The following reply was made to PR kern/92518; it has been noted by GNATS.
From: phillip at it.jcu.edu.au
To: bug-followup at freebsd.org
Cc: phillip at it.jcu.edu.au
Subject: Re: kern/92518: Intense disk activity (large FS newfs and/or many small files copy): DMA completion error with RocketRAID-1820A and FastTRAK-S150
Date: Thu, 16 Feb 2006 22:37:43 +1000
The following work-around for this bug has proved successful:
1) use the RocketRAID for a RAID-1 arrays constructed on pairs of disks.
Assuming same sized disks, you'll end up with same sized RAID-1 arrays.
2) instead of using the RocketRAID drivers for striping to produce a
RAID-10 array, use a GEOM stripe across the RAID-1 arrays.
This allows newfs of 240GB file systems with no problems, and test copies
of large numbers of small files (such as are in /usr/ports and /usr/src)
also works fine.
In addition, the RocketRAID driver can still manage on-line spare disks in
the underlying RAID-1 arrays.
The speculation that led to this work-around was that perhaps the EDMA time
outs are being caused by events being confused between the RocketRAID
software drivers associated with RAID-1 and RAID-0. Having RocketRAID
handle RAID-1 and GEOM handle higher levels gets around this. However, I
have not tested this speculation.
The RocketRAID RAID-1 + GEOM stripe arrays have been in normal use for
about almost week. It looks good.
Note system update: the system has gone from 6.0-STABLE to 6.1-PRERELEASE
during which the RocketRAID driver has not changed.
More information about the freebsd-bugs
mailing list