HPE Storage Users Group

Hi there guys,

According to some folks at HP, when adding a new drive type to a 3PAR, this must be done in groups of 8 drives, just wondering: Is this because of a recommended "best practice" or its actually a limitation? Meaning, a 3PAR will simply ignore a group of less than 8 disks added?

I ask because some people need just a small quantity of SSD drives and adding this drives in 8´s makes 3PAR kinda expensive.

Thanks in advance for any comment!

How many trays I heard they prefer minimum 2 drives per tray as min disk config. So assuming a 2 tray install that would sound like 4 drives.

I've had similar questions and never was able to get to the bottom of it... I just installed in sets of 8.

2 per node pair would only support raid 0. (1 owned by each node, nothing left to mirror to)
4 per node pair would only support raid 0/1. (2 owned by each node)
8 per node pair would enable R5 3+1, the flagship standard for 3PAR. (each node would have 4 drives to perform a 3+1 set on).

_________________
Richard Siemers
The views and opinions expressed are my own and do not necessarily reflect those of my employer.

My interpretation of it is something like this (this only applies to 7200's, and specifically, I'm thinking about one of my 7200's which has 4 shelves (including the controller enclosure) of SFF drives.)

You must install drives in pairs (HP states this in quickspecs)
In order for backend IO to be well balanced, each tray should have the same number of drives (this is also necessary for cage-level availability)

This would make the minimum upgrade quantity on my 7200 8 (2 drives per tray * 4 trays).

Additionally (and it probably isn't necessary on 3PAR, but I like to keep it this way regardless), you would want the number of drives to be a multiple of your RAID set size.

So if you're running R5 3+1, then you would want drives in multiples of 4 so that your raid aligns with your drive count.

If I were to install SSD's, I would install 8, and put them into a R5 3+1 CPG with cage level availability (2 per shelf).

If you had a 7200 with only 2 shelves, you could probably get away with installing 4

Richard makes a good point though about the nodes owning drives.

Perhaps a call to HP can clear things up and we can all learn a few things?

Eight drives are required the build a minimum 3+1 stripe across two controllers (A/A), anything less and you would be limited to Raid10. But there are also minimum disk requirements for any new class of disk as metadata (SA space) needs to be triple mirrored. So your absolute theoretical minimum would be six to support triple mirroring, but I have no idea whether that would actually work for user (SD space). Even if it did, the best you could do would be Raid 10, so better to add 2 more drives and get much more flexibility and capacity.

Minimum upgrades are pairs, one per controller but this assumes your happy with mag level availability. If you want cage and assuming you have the enclosures, then the quantity needs to be equally divisible by the stripe size. e.g 2 controllers x 3+1 = 8, 16, 24, 32 etc if not pairs are fine, but minimum 8 for a new class of disk.

Thanks guys, I spoke with an HP representative, he wasn't able to confirm if less than 8 drives would work in the system, or why that works in that way. However, he was very clear telling me that this is not supported. I guess we'll never know. Cleanur explanation seems to make sense though.

Indeed. Cleanur's explanation makes sense.

Anyway, if you use the 920GB SSD drives, you'll need 8 at least to get any usable capacity. At r5 3+1, you'll only get like 5 tb usable after raid and all that and it goes quick

With 7xx0 drives must be added by pair.
If you got 3 SFF cages you must add 6 drives minimum.
With 7000 class ha cage is no longer the classic ha.

The question was about adding a new drive type, for which the rule is that a minimum of 8 drives must be installed, no matter how many shelves there are, that's the minimum.

Also, theres nothing that says drive types must be evenly distributed among cages. For example, I have 4 sff cages. There is nothing stopping me from putting 4 ssd in cage 0 and 4 in cage 1 if I were so inclined. The only time that would be required is to meet ha cage set size requirements