On Sun, Oct 14, 2012 at 02:33:32PM +, Thor Lancelot Simon wrote:
Module Name: src
Committed By: tls
Date: Sun Oct 14 14:33:32 UTC 2012
Modified Files:
src/sys/ufs/ufs [tls-maxphys]: ufs_readwrite.c
Log Message:
In the FFS writebehind code (ufs_readwrite.c:WRITE()), use division
and multiplication instead of shifts, to accomodate devices with
MAXPHYS that is a multiple of the page size, but not a power of 2.
MegaRAID neatly writes out 192k chunks now.
An open question: is is really a good idea to always writebehind
at the largest size supported by the device? Likely not, as this
could have a major impact on I/O fairness. OS X and Solaris both
seem to limit transfers to 128k likely for this reason (the same
problem exists with the readahead code but since it is adaptive,
it will not *always* do huge transfers).
However, simply imposing a smallish limit like 128k here seems
like a bad idea because then we cannot accomodate greedy devices
like RAID, for which you want something like 128k * number of
data components. Hmm.
I wonder if we could reuse bits from the read-ahead code for write-behind ?
Maybe this could allow to move the write-behind code out of ufs_readwrite.c
to ubc (maybe ubc_uiomove()) ?
--
Manuel Bouyer bou...@antioche.eu.org
NetBSD: 26 ans d'experience feront toujours la difference
--