large RAID volume partition strategy

[Kirill Ponomarew]

On Wed, Aug 29, 2007 at 10:07:19AM -0400, Vivek Khera wrote:
> 
> On Aug 17, 2007, at 10:44 PM, Ivan Voras wrote:
> 
>> 
>> fdisk and bsdlabels both have a limit: because of the way they store the
>> data about the disk space they span, they can't store values that
>> reference space > 2 TB. In particular, every partition must start at an
>> offset <= 2 TB, and cannot be larger than 2 TB.
> 
> Thanks.  This is good advice (along with your other note about doing it in 
> the RAID volume manager).  Nearly everyone else decided to jump on the raid 
> level instead and spew forth the "RAID10 is better for database" party 
> line.  Well to you folks: once you have 1Gb cache and a lot of disks, there 
> is not much difference between RAID10 and RAID5 or RAID6 in my testing.
 
What type I/O did you test, random read/writes, sequential writes ?
The performance of RAID group always depends on what software you
run on your RAID group.  If it's database, be prepared for many
random read/writes, hence dd(1) tests would be useless.

> I ended up making 6 RAID volumes across all the disks to maximize spindle 
> counts and strip the data at 16kB.  This seems to work well, and I can 
> assign the other partition as I need later on.
> 
> _______________________________________________
> freebsd-stable at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-stable
> To unsubscribe, send any mail to "freebsd-stable-unsubscribe at freebsd.org"

-Kirill