We solved a similar problem by limiting the buffering per device/partition to
a subset of the buffer cache (75%, tunable), rather than hacking in raw
devices.
This worked extremely well for our application, a parallel file system.
The system had following features:
Multiple disk partitions. (Note 1)
Bulk I/O tended to be on just one partition at a time.
Most other processes were active in partitions without bulk I/O.
Note 1: This approach has no benefit on single-partition installations.
However most bulk I/O users do have multiple partitions, and typically have
multiple disks.
Donald Becker
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/