Confusion about EVA RSS Disk State

[Joe Hegyi]

I'm still a bit confused as to how I can fix my RSS Disk State problem. After adding 2 additional shelves and disks, our RSS state went from Mirrored to None. I would like to know if doing either of the following would get it back to Mirrored.

Current EVA 5000 setup (A = Pool 1, B = Pool 2, x is an empty drive bay)

Shelf 11) AAAAxxxBBxxxxx (4A, 2B)
Shelf 10) AAAAAAABBBBBBB (7A, 7B)
Shelf 09) AAAAAAABBBBBBB (7A, 7B)
Shelf 08) AAAAAABBBBBBBB (6A, 8B)
.
.
.
Shelf 04) AAAAxxxBBxxxxx (4A, 2B)
Shelf 03) AAAAAAABBBBBBB (7A, 7B)
Shelf 02) AAAAAAABBBBBBB (7A, 7B)
Shelf 01) AAAAAABBBBBBBB (6A, 8B)
---------------------------------

*Proposal 1*
Shelf 11) AAAxxxxBBBxxxx (3A, 3B)
Shelf 10) AAAAAAABBBBBBB (7A, 7B)
Shelf 09) AAAAAAABBBBBBB (7A, 7B)
Shelf 08) AAAAAAABBBBBBB (7A, 7B)
.
.
.
Shelf 04) AAAxxxxBBBxxxx (3A, 3B)
Shelf 03) AAAAAAABBBBBBB (7A, 7B)
Shelf 02) AAAAAAABBBBBBB (7A, 7B)
Shelf 01) AAAAAAABBBBBBB (7A, 7B)
--------------------------------
*Proposal 2*
Shelf 11) AAAAAAABBBBBBB (7A, 7B)
Shelf 10) AAAAAAABBBBBBB (7A, 7B)
Shelf 09) AAAAAAABBBBBBB (7A, 7B)
Shelf 08) AAAAAAABBBBBBB (7A, 7B)
.
.
.
Shelf 04) AAAAAAABBBBBBB (7A, 7B)
Shelf 03) AAAAAAABBBBBBB (7A, 7B)
Shelf 02) AAAAAAABBBBBBB (7A, 7B)
Shelf 01) AAAAAAABBBBBBB (7A, 7B)

Would either proposal 1 or proposal 2 (which basically fills all 8 shelves and gets the pools lined up) automatically set the RSS Disk State to Mirrored?

Thanks,
Joe

[Matt Burgess]

This post is long, so I have attached a txt file to make it easier to read.

**

So.

Old config = 2 disk groups (40 HDD, 44 HDD), 6 enclosures
New Config = 2 disk groups (48 HDD, 48 HDD), 8 enclosures


This is the best way to place the drives:

11)xAAABBBBBBAAAx
10)xAAABBBBBBAAAx
09)xAAABBBBBBAAAx
08)xAAABBBBBBAAAx

04)xAAABBBBBBAAAx
03)xAAABBBBBBAAAx
02)xAAABBBBBBAAAx
01)xAAABBBBBBAAAx

but this won't automatically fix up your RRSs. I'll explain.

Your RSSs are confured based on the best configuartion on the day they are created. What this means is if you add enclosues or physically move disks, the 'best' configuration from day 1 isn't the best anymore.

If you look at the RSSs (CV EVA, go to the disk and locate the RSS) they might look something like this currently. I'll label the RSSs 1-9 and A-C (There should be 11 in the old config, 12 in the new)

RSS 1-5 are Disk Group 1 (Each have 8 disks)
RSS 6-A are Disk Group 2 (6-9 have 8, A and B have 6)

Your old config (11 RSS):

11)
10)12345346789A9B
09)12345346789A9B
08)12345266789A8B

04)
03)12345256789A8B
02)12345156789A7B
01)12345166789A7B

I can't guarentee it will look like this, but the point here is that you will find more than one disk from each RSS on each enclosure. There will also be mirrored pairs fomr the VRAID1s that are on the same shelf, potentially.

You new Config (12 RSS)

11)CCCCxxxABxxxxx
10)12345346789A9B
09)12345346789A9B
08)12345266789A8B

04)CCCCxxxABxxxxx
03)12345256789A8B
02)12345156789A7B
01)12345166789A7B

When you add enclosures and disks, it doesn't reconfigure existing ones, it just creates new ones (like RSS C) and/or adds the members to existing ones to get the number of members closest to 8 (like RSS A and B in this example)

The thing to understand here is that an RSS state of "Mirrored" is not the best. It is broken down.
None= enclosure failure --> disk group will be Offline
Mirrored=enclosure failure --> VRAID5 VDISKS will be Offline, VRAID1 VDISKs are OK.
Parity=enclosure failure --> VRAID5 and VRAID1 VDISKS are ok.

SO you don't want "Mirrored". You want "Parity" You want to end up with this
Remembering DG1 RRSs=1,2,3,4,5,C and DG2 RSSs=6,7,8,9,A,B:

11)x123678 9ABC45x
10)x123678 9ABC45x
09)x123678 9ABC45x
08)x123678 9ABC45x

04)x123678 9ABC45x
03)x123678 9ABC45x
02)x123678 9ABC45x
01)x123678 9ABC45x

The reason for this is because you may lose 1 member from each RSS in

The process for this is:

1. Locate your RSSs. Draw a map of them.
2. Draw a map of what you want to end up with, like I just did.
3. Go to the Storage System in CV EVA and change the 'disk replacement delay' to 3 minutes. This is the timer that tells when the EVA should treat a 'missing' disk as a serious error.
4. Move each disk from its old location to its new location one at a time.
-this means physically remove it (DO NOT UNGROUP)
-place it in its new location
-wait for all errors to go away and leveling % to be N/A
-now you can move the next disk

Aside from this.
5. If any of the RSSs have a number of disks more than 8, UNGROUP the members. When leveling has completed, you can add them back to the group. Example. RSS1=6disk, RSS2=10 disks. If you ungroup 2 disks from RSS2 and then add them back in to the disk group, the new 'best' configuration will feature 2 RSSs of 8, so it will add the 2 disks to RSS1.

The guidelines for Disk placement (See very first diagram in post)
DISK GROUP.
1. Distribute Disks accross different enclosures.
2. Distribute Disks accross different sides of the enclosure (bays 1-7) (bays 8-14)
3. Disk group should have #of disk divisible by 8.

RSS COnfiguration Algorithms used by HSV
1. Different enclosures, if possible, if not then,
2. Different sides of the enclosure
3. 8 is the best number, then 6, then whatever is left over (7,9,10 and 11 are acceptable) (12 will create 2x6)

I will also attach this post as a text file and subscribe for notifications.

cheers

Matt Burgess