Re: Huge Data sets, simple queries

On Tue, Jan 31, 2006 at 12:47:10PM -0800, Luke Lonergan wrote:
>> Linux does balanced reads on software
>> mirrors. I'm not sure why you think this can't improve bandwidth. It
>> does improve streaming bandwidth as long as the platter STR is more than
>> the bus STR.
> ... Prove it.

FWIW, this is on Ultra160 disks (Seagate 10000rpm) on a dual Opteron running
Linux 2.6.14.3:

cassarossa:~# grep md1 /proc/mdstat
md1 : active raid1 sdf6[1] sda6[0]
cassarossa:~# dd if=/dev/sda6 of=/dev/null bs=8k count=400000
[system at about 35% wait for I/O and 15% system, according to top]
400000+0 records in
400000+0 records out
3276800000 bytes transferred in 54,488154 seconds (60137842 bytes/sec)
[system at about 45% wait for I/O and 7% system -- whoa?]
400000+0 records in
400000+0 records out
3276800000 bytes transferred in 52,523771 seconds (62386990 bytes/sec)

I'm not sure if it _refutes_ the assertion that the Linux RAID-1 driver can
do balancing of sequential reads, but it certainly doesn't present very much
evidence in that direction. BTW, sda and sdf are on different channels of a
dual-channel (onboard, connected via PCI-X) Adaptec board, so I doubt the bus
is the limiting factor.

/* Steinar */
--
Homepage: http://www.sesse.net/