RAID6 - whole disk vs partitions

RAID6 - whole disk vs partitions

[Steven Haigh]

[-- Attachment #1: Type: text/plain, Size: 1161 bytes --]

Hi all,

I'm revisiting this topic after a few years - so excuse me if I'm a
little behind the times on this questioning.

I'm just starting to replace some 5 year old disks in a RAID6 (5 disks)
due to one having started to throw uncorrectable / reallocated sectors.
As such, it seems a good time to go to 3Tb drives - which are only $20
more than 2Tb drives.

The array is created as follows:
md2 : active raid6 sdf[0] sde[6] sdg[7] sdd[5] sdc[8]
      5860150272 blocks super 1.2 level 6, 128k chunk, algorithm 2 [5/5]
[UUUUU]

I've always used the whole disk for the RAID6 - as from what I
understand, this gives the best layout for various drives - no matter if
they're 512 or 4096 byte sectors.

Is this still the case?

I've noticed that my resync speeds on this RAID6 hover around 60Mb/sec -
which I was expecting to be a bit quicker (although I haven't tested
speeds on this array for years).

Currently using kernel 4.1.15 on an EL6 install.

Would be grateful for any input people may have on these topics.

-- 
Steven Haigh

Email: netwiz@crc.id.au
Web: https://www.crc.id.au
Phone: (03) 9001 6090 - 0412 935 897


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

Re: RAID6 - whole disk vs partitions

[Phil Turmel]

Hi Steven,

On 12/19/2015 09:34 AM, Steven Haigh wrote:

> I've always used the whole disk for the RAID6 - as from what I
> understand, this gives the best layout for various drives - no matter if
> they're 512 or 4096 byte sectors.
> 
> Is this still the case?

There's no real advantage other than an itty bit of space.  If my case
was too small to have a separate mirror for my OS, I'd make three
partitions on everything for boot, LVM root & etc (raid10), and bulk
storage (raid6).  When I have room for the mirrored OS (boot & LVM), I
use whole disks for the bulk storage.

> I've noticed that my resync speeds on this RAID6 hover around 60Mb/sec -
> which I was expecting to be a bit quicker (although I haven't tested
> speeds on this array for years).

Consider using small chunk sizes for parity raid if there's even the
slightest random access usage.  See the list archives for the past
couple weeks for a long thread on performance.  There's many
optimization possibilities.

Phil

Re: RAID6 - whole disk vs partitions

[Steven Haigh]

[-- Attachment #1: Type: text/plain, Size: 1854 bytes --]

On 20/12/15 07:36, Phil Turmel wrote:
> Hi Steven,
> 
> On 12/19/2015 09:34 AM, Steven Haigh wrote:
> 
>> I've always used the whole disk for the RAID6 - as from what I
>> understand, this gives the best layout for various drives - no matter if
>> they're 512 or 4096 byte sectors.
>>
>> Is this still the case?
> 
> There's no real advantage other than an itty bit of space.  If my case
> was too small to have a separate mirror for my OS, I'd make three
> partitions on everything for boot, LVM root & etc (raid10), and bulk
> storage (raid6).  When I have room for the mirrored OS (boot & LVM), I
> use whole disks for the bulk storage.

Yes, I have two separate disks in RAID1 as the boot / VM guests - the
RAID6 is used for bulk storage. The chassis has a total of 10 hot swap
bays, with 5 used by the RAID6, 2 by the RAID1 and 3 spare. I get resync
speeds of around 170Mb/sec on the RAID1.

>> I've noticed that my resync speeds on this RAID6 hover around 60Mb/sec -
>> which I was expecting to be a bit quicker (although I haven't tested
>> speeds on this array for years).
> 
> Consider using small chunk sizes for parity raid if there's even the
> slightest random access usage.  See the list archives for the past
> couple weeks for a long thread on performance.  There's many
> optimization possibilities.

I have tried to keep up with those couple of threads - but I've found it
hard to follow across the length of time they've been going. A summary
of the findings in those threads with experimentation would be fantastic!

It seems that actual data beyond "increase --setra on /dev/sd[.] and
increase the stripe_cache_size is about all I find via Google - and that
data is years old...

-- 
Steven Haigh

Email: netwiz@crc.id.au
Web: https://www.crc.id.au
Phone: (03) 9001 6090 - 0412 935 897


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]