Start a Conversation

Unsolved

This post is more than 5 years old

A

5 Practitioner

 • 

274.2K Posts

5921

September 15th, 2010 05:00

How will Flare handle lun in a thin pool , when expanding with more disks

Let's assume we have 20 disk in 4 x 4+1 R5

and I assume that when creating a lun , let's say 100 GB init. it will be distributed on all 20 disk


When customer then whants to add more disk , will that extent the exsisten lun out to the 25 disk or do they have to do a lun migration , on by on


Thanks

John

50 Posts

September 24th, 2010 09:00

John,

What version of Flare are you running in this scenario, Thin pools have been around since flare 28xxxxx.  Below flare 30 there is no auto load balancing of a LUN within a Thin Pool.  When you create the LUN in a Thin Pool its thinly allocated so you provision a 100gb LUN but the pool allocates just 1gb LUN initally.  Since your using 20 disks in a raid 5 Thin Pool(TP), the TP has created in the background 4+1 for every 5 of your 20 disks.  So first 1 gb slice is from the first 4+1 in the pool   As your LUN begins to use space the pool allocates the next 1gb slice from the next 4+1 and so on and so on where it eventually wraps back around.  Your LUN is now a bunch of slices from the pool but presented as a single LUN.  So if you were to add disks later on to the pool and you wanted to relevel and you had the space, LUN migration within the same pool would be needed.

IN flare 30, be sure to get Fast 2 because it auto levels these individual 1gb slices.  Your LUN can now live on EFD, FC and SATA based on needed performance.  So releveling is not so much concern any more.  Definately read up on Thin Pool allocation and Fast 2 becuase there is huge benefits there.

Thanks,

Mike

September 28th, 2010 13:00

Mike, your reply was a little vague. You say that we should read up on FAST 2, which is fine; however, are you saying that with R30, and FAST 2, you get auto-leveling within the same drive type, but not without FAST 2?

For example, if I have R30, and I have 20 FC drives in a pool, then I add 5 more drives, walk me through where the data goes. If I add 100GB of data, which spindles does that 100GB go to? And, what happens to the 2TB of data that was on the original 20 drives?

I would like to see more transparency about Storage Pools from EMC. More white papers, more scenario based discussions, etc. Those of us that have lived in the real (non virtual, non thin) world for many years are hesitant to throw out all our best practices and rely on the array to figure it out for us. We need to know more details about where our data is so that we can prevent hotspots and better tune our environment. Before I start throwing SATA drives in an FC pool, I'm going to want to feel comfortable expanding FC with FC - which I hope does not require FAST to get full benefits. I read in one of EMC's detailed reviews that it is "recommended" to expand thin pools with the same number of disks you start out with. To me, this is sounds like MetaLUNs - striped vs concatenation - and needs more explanation.

Thank you,

Jason Burroughs

474 Posts

November 4th, 2010 12:00

Jason,

If you don't have FAST features in FLARE30, then the answer is pretty much the same for FLARE28,29,30..  Adding disks to a pool will not rebalance existing data.  So if you created a pool of 30 disks and 10 LUNs, and wrote data for some period of time, then added 30 more disks to the pool, the existing data will remain on the first 30 disks, only new data will be written to the new disks.  That new data would include the existing LUNs as well as any new LUNs created after pool expansion.  But existing data will never be moved.  There will be performance improvement but it will not be linear as it would be in VMAX after a pool-rebalance operation.

With FASTv2, subLUN-FAST, FASTVP (the name depends on when you heard about it and who are talking to) which is optional in FLARE30, pools are created using different types of disk (EFD/SSD, FC, SATA) and data is moved between those disk types (tiers) automatically based on IO loads measured in 1GB slices.  Since this movement happens fairly regularly, the pools sort of self-balance over time.

For non-FAST environments, if you need to rebalance a pool because you added disks and are experience a performance problem, you can use LUN migration to migrate an old LUN to a new LUN in the same or different pool.

1 Message

March 6th, 2011 20:00

I did some testing on storage pools. Here are my findings: http://virtualeverything.wordpress.com/2011/03/05/emc-storage-pool-deep-dive-design-considerations-caveats/

Appreciate any feedback/corrections/etc.

Thanks.

/vijay

No Events found!

Top