On Fri, Oct 04, 2013 at 04:15:22PM +0000, ray clancy
wrote:> How can I verify the read speed of a btrfs raid0 pair in archlinux.?
>
> I assume raid0 means striped activity in a paralleled mode at lease
> similar to raid0 in mdadm.
>
> How can I measure the btrfs read speed since it is copy-on-write
> which is not the norm in mdadm raid0.?
Testing read speed... you''re not writing, so there''s no
copy-on-write involved there. Just test reading the way you would for
anything else.
> Perhaps I cannot use the same approach in btrfs to determine the
> performance.
>
> Secondly, I see a methodology for raid10 using the
> command....mkfs.btrfs -m raid10 -d raid10 /dev/sda/dev/sdb /dev/sdc
> /dev/sdd...
>
> Can I apply the parameters above for -m and -d for raid0?
I''d certainly recommend it for testing RAID-0. :)
Actually, a slightly more realistic test would be to use RAID-0 for
data and RAID-1 for metadata, because that''s what most [default] users
of the FS will end up with.
> If using raid0 for two devices and add another device, is it striped
> as raid0 also or does the system change it to raid1.
No, it''ll remain as RAID-0. If you rebalance, then the data will
get striped across three devices instead of two.
> What happens to the speed of the system when a new device is added?
> Is it increased ?
Assuming the FS is reading from all the devices, yes.
> Much I have at hand for mdadm software raido and it doubles the read
> speed. What parallel exists in raid0 btrfs? Or is it completely
> off base to expect a speed increase?
In theory, you should be able to get the sum of the bandwidths of
all the devices (assuming sequential streaming reads). We don''t have
any good benchmarks of this kind of thing, so when you do your tests,
please (a) make sure you do a decent experimental design, and (b)
publish the results. :)
Hugo.
--
=== Hugo Mills: hugo@... carfax.org.uk | darksatanic.net | lug.org.uk == PGP
key: 65E74AC0 from wwwkeys.eu.pgp.net or http://www.carfax.org.uk
--- I always felt that as a C programmer, I ---
was becoming typecast.