EVA8100 Levelling

[Paul Hawkins]

I have recently become responsible for an EVA8100. I have just added 4 disks into the default diskgroup and created a couple of vdisks that were desperately required by a project. The array is just about full (Occupancy level of 33395 v capacity of 36869 GB)soo I would expect levelling to take a while.
My question is that what would the impact be if I change the diskdrive failure protection policy to "single" from "double".....
Would this affect the time that it takes to complete?

[IBaltay]

Hi,
you will free the capacity of the 2 virtual disk drive to be used in the DG. If there is more data uin the DG - also the back-end metadata operations - leveling should go faster.

[Víctor Cespón]

The "disk failure protection" reserves spare space to be able to rebuild the array in case of a disk failure.
Single reserves 2 x (size of largest disk on disk group). Double reserves 4 x (size of largest disk on disk group).

Switching to Single will make 2 x (size of largest disk on disk group) available as space on the disk group.

Double is only recommended in extreme cases where you have 200 disks or so on a disk group, or the EVA is in a place where a replacement disk cannot arrive in a few hours.

The probability of 2 disks on the same disk group failing in a few hours is too low.

[Colin Butcher]

Right now you probably would slow down the levelling process if you changed the sparing policy.

Given how big the EVA is and the number of discs you've probably got, why not just fill it up completely with spindles, thus reducing the occupancy levels a bit and set up double sparing? While you're at it do think about the RSS groups and how you're grouping drives up across the shelves. Ideally you want your disc groups made up of blocks of 8 drives, where within each group of 8 there's one drive per shelf.

Sit down with the "best practice" white papers and start reading.

Using SSSU to list everything out, then mangling that text output with some PERL scripts can be used to create CSV files, which you can then look at easily in table format.

Once you have a picture of how the EVA is set up, then you can start to think about what makes sense to change.

If it's that full, hopefully you have a second one someplace with data replication (VMS HBVS, CA, whatever) between them.

Cheers, Colin (http://www.xdelta.co.uk).

[McCready]

The disk drive failure level is basically reserved space to be used to reconstruct the drive data from the parity contained within the RSS group - you could have the same effecgt by simply not allocating that amount of space (but then, you will get people wanting to use it).

Specifically, I'm not sure if it really will affect leveling speed, but generally, the more space you give an EVA, the better it performs. So I would agree on not wanting to reduce the available space by changing the policy to double.

Overall, what you most want to minimize is the time it takes to reconstruct the data from parity - you get a warning pop-up until that is done from command view when looking at the EVA. While that pop-up/reconstruction is happening, you are at risk of data loss if you lose another disk (raid 5) or the "mirror" disk (raid 1) from the same rss group.

Leveling time is also affected by the I/O rate on your EVA, as leveling I/O's will wait for host I/O's to be done.