HPE Storage Users Group

We have 3PAR 7200 with 28x HDD.
Can I config CPG with RAID5(8+1)?
Will it make impact on performance or usable capacity?
Or I can only config CPG with RAID5(6+1)?

I only have 2 enclosure (including node enclosures),and I don't care about cage resilience.
Someone say that the disk count must be dividable by RAID type.
28HDD should be config with RAID5(6+1),
If I config with RAID5(7+1) or (8+1), It will waste space.
Is that true?

Posheng wrote:We have 3PAR 7200 with 28x HDD.
Can I config CPG with RAID5(8+1)?
Will it make impact on performance or usable capacity?
Or I can only config CPG with RAID5(6+1)?

How many enclosures do you have? My understanding of best practice is that if you run HA Cage you want your RAID stripe to be dividable by your enclosure count (including node enclosures) if possible - IE, 3+1 on 4 enclosures, 7+1 on 8 enclosures, 4+1 on 10 enclosures ETC.

If you just have the node enclosure + 1 or 2 disk enclosure you can't run HA cage in R5.

Your limiting factor will be only 28 HDDs more than your R5 ratio. We have had pretty good luck with 7+1 as far as performance fwiw.

As above, please let us know how many shelves these 28 disks are spread over so that we can let you know whether cage resilience is available.

If you don't care about cage resilience then go for whichever config you think, I'm led to believe that increasing the ratio (I.e. 6+1 instead of 3+1) doesn't really have an impact on speed, but will give you more space.

I only have 2 enclosure (including node enclosures),and I don't care about cage resilience.
Someone say that the disk count must be dividable by RAID type.
28HDD should be config with RAID5(6+1),
If I config with RAID5(7+1) or (8+1), It will waste space.
Is that true?

Yeah it will mess with the fengshui if you do R5(7+1) or (8+1) with 28 drives. Sweet spots for 28 drives will be R1(1+1) HA CAGE, R5(6+1) HA MAG, and possibly R6(12+2) HA MAG if it meets your application needs.

With 2 nodes, you will have 2 pools of 14 disks for each node to "make" raid Logical Disks from. "LD" creation and growth is a deep dive technical read that I myself haven't fully grasped the entire concept of yet. If you really want to dive into the how and why, check out "Logical Disks" in the Concepts Guide.

If you aren't bothered about cage, 28 disks total, 14 disks per node. So long as you stripe size is lower than the total disks per node then you're good to go.

If the stripe size isn't a a factor of 14 then dependent on the number of rows employed you have the potential to have a small amount of unused space (we're talking a few GB's of chunklets per node) where the array can't form the last stripe as there aren't enough remaining chunklets. See simplified example below, which represents chunklets on disk creating raidlets which form LD's and eventually volumes.

The smaller the stripe the less overhead for parity calculations and subsequent updates but you'll probably only see a difference if you are really pushing the array which won't be possible with so few disks. Obviously the wider the stripe the more efficient as there's less parity overhead 2+1(33%), 3+1(25%), 4+1, (20%), 5+1 (17%), 6+1(15%), 7+1(13%), 8+1 (12%). But the wider the stripe then mathematically, the higher the likelihood of a dual disk failure within that stripe.

So your optimal would be 6+1 (2x7x2nodes) on Raid 5 that way you can in theory fill every chunklet. Remember raid is built at a chunklet not a disk level so don't get too hung up about it, each drive can house multiple raid type and stripe sizes. That's why we have CPG's and unless you're thick provisioned you'll likely never fill them all anyway.

Nice explanation cleanur

Thank you! I think I know how to config it~