HPE Storage Users Group
https://3parug.com/

Critique this disk layout, please.
https://3parug.com/viewtopic.php?f=18&t=3375
Page 1 of 1

Author:  richard612 [ Thu Mar 19, 2020 11:17 am ]
Post subject:  Critique this disk layout, please.

What we have:

* 4 node 7400 running 3.2.2
* 20 shelves
* 8x 200GB SLC SSD
* 8x 400GB MLC SSD
* 8x 920GB MLC SSD
* 214 1.2TB drives
* Nearly unlimited supply of 900GB drives (long story).

I'm trying to configure for max capacity, availability=cage across the entire system, and optimal use of the SSD as part of an AO config. This is what I've come up with so far. Having only 8x of each SSD type is a real drag. I'm wondering if "express layout" could be helpful here and how to leverage it. Any thoughts or comments would be very appreciated.

Attachment:
Capture2.png
Capture2.png [ 419.98 KiB | Viewed 14746 times ]

Author:  MammaGutt [ Thu Mar 19, 2020 1:15 pm ]
Post subject:  Re: Critique this disk layout, please.

You have (as you seem to know) one problem.

The low number of SSDs.

Just a few pointers.

Ensure that each node pair is as close to equal as possible (which you seem to have).

Express layout allows both nodes in the node pair to access all drives, or in other words, one drive will serve both nodes in the pair.

If you run R5 (3+1) on SSD CPG you will be able to fully allocate the SSDs if you have Express Layout (3.2.2 MU2 and above).

To me it looks as good as it gets with what you have. The only thing that might make an impact when you get to 90+% is the last cage with 10x900GB when you have others with 14.

Author:  richard612 [ Thu Mar 19, 2020 1:50 pm ]
Post subject:  Re: Critique this disk layout, please.

Thanks for your reply. I've begun to implement this and for some reason it says available space is zero when trying to create a 2+1 or 3+1 RAID5 cpg. I've tried it against all three SSD sizes and even with availability set to magazine. It's only letting me create RAID1 with 2, 3, or 4 data. I don't get it. I kicked off a tunesys but given how many disks I've moved around it's going to take a month.

The express layout verbiage from HPE suggests one SSD per shelf is sufficient and can be shared across both nodes. With that in mind i wanted to try putting the SLC in slot 0 across shelves 0-7 and see what happens. If it works and checkhealth is happy I'll repeat the process for the MLC disk. That layout would be much uglier...

(TLC is the 920GB disk - it's actually MLC and I need to fix the spreadsheet)

Attachment:
Capture2_2.png
Capture2_2.png [ 431.67 KiB | Viewed 14736 times ]

Author:  richard612 [ Thu Mar 19, 2020 1:54 pm ]
Post subject:  Re: Critique this disk layout, please.

Also, I was planning to run the 900s and the 1200s together in the same CPG. The loss of IOPS when the 900s fill up isn't too worrisome (assuming we ever fill it that far). Anything performance intensive or performance sensitive can go onto our Nimble AF unit. As for the 3PAR flash, we could live without it but since it's here I want to try to use it as best I can.

Author:  MammaGutt [ Thu Mar 19, 2020 4:06 pm ]
Post subject:  Re: Critique this disk layout, please.

Drives shall be installed in pair, so single SSD per cage is bad.

Even if it is "technically OK" for SSDs with Express Layout, the FC drives doesn't do EL on 3.2.2.

Remember you should have one SSD CPG covering all SSDs in your config.

Yeah, i forgot your cage HA... you have 12 SSDs per node pair so should should get SSDs in 6 cages per node pair.

What is your exact 3PAR OS version?

As for 900/1200 drives in the same CPG you have so many 1200 disks I doubt you will even notice it.

Author:  richard612 [ Thu Mar 19, 2020 4:20 pm ]
Post subject:  Re: Critique this disk layout, please.

Quote:
"Remember you should have one SSD CPG covering all SSDs in your config"

In that case I'd probably be fine with my first plan. At any rate I've already cleared off the SLC drives in column 1 and the 900GB interlopers in column 0. Next trip to the datacenter I'll toss the SLCs into slot 0 across cage 0-7 and see what it'll let me do and what checkhealth says. Worst case I'll put them back where they were.

We could probably live without availability=cage which would open a lot of placement possibilities but I'm gonna fight to keep that setting for now.

Page 1 of 1 All times are UTC - 5 hours
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group
http://www.phpbb.com/