Start a Conversation

Unsolved

This post is more than 5 years old

3284

September 8th, 2012 09:00

Pool LUN overhead

Hi All!

I`ve read EMC VMX Virtual Provisioning White Paper and foud out that initial overhead is 3 GB. Some of those 3 GB used for metadata and remaining capacity ready for incoming writes. Ful overhead for whole LUN = LUN_size*0.02 +3 GB.

For example i need 100 GB pool LUN. This LUN will take 103 GB from pool. Am I correct?

So why does initial 3 GB not included in the LUN size? White papre says that some of the 3 GB is used for metada. But overhead is 3 GB too. I don`t understand, please help me :-)

5.7K Posts

September 13th, 2012 02:00

Then please explain why I had 7GB available in a pool and after I cteated a 1GB LUN I only had 3GB left and I wasn’t able to create another 1GB LUN.

247 Posts

September 13th, 2012 02:00

First of all, if you make a 100GB LUN, you want to assign 100GB of space to a server. It would be very confusing if the VNX would just grab the overhead from that 100GB LUN, leaving you with a 97GB LUN. It would make sizing more of a trial and error process...

The actual situation is real simple. You create a lun, say 100GB. On top comes xGB of metadata + 2% of extra overhead. That's it!

Graphically, you'd see:

PoolLUN.JPG

This is a 1024GB / 1TB LUN from server perspective, but on the actual pool it consumes a bit more. The overhead is just the Consumed Capacity - User Capacity.

247 Posts

September 13th, 2012 03:00

But that's simple RRR:

Start 7GB free in the pool.

I create a 1GB LUN. In reality, it uses 1GB for user space, ~3GB for overhead.. Total usage = 4GB.

Pool free: 7-4=3.

If i would want to create another 1GB LUN, i'd need again 1GB of user space, 3GB of overhead. Which is 4GB, which doesn't fit in a pool that only has 3GB free. Error!

5.7K Posts

September 13th, 2012 05:00

That’s a lot of overhead, percentage wise that is.

247 Posts

September 13th, 2012 06:00

Yeah, but let's be honest... how often do you allocate 1GB LUNs? I think 9 out of 10 LUNs are 1TB over here, for the VMware farm. The rest are some physical servers that use some storage; LUN sizes range from 10 to 600GB.

5.7K Posts

September 13th, 2012 06:00

Almost never, but in this config I had to create 25TB worth of LUNs and I left the 1GB LUNs for last. Don’t ask me what they were for, could be quorum disks, I don’t remember.

Van: Jon Klaus

Verzonden: donderdag 13 september 2012 15:15

Aan: Rob Koper

Onderwerp: New message: "Pool LUN overhead"

EMC Community Network

Re: Pool LUN overhead

created by Jon Klaus in VNX Support Forum - View the full discussion

16 Posts

September 13th, 2012 08:00

Jon, thank you.

I understand the differences betwen user capacity and consumed capacity. But i can`t understand why does a very small lun has so huge amount of metadata. It is not a problem for me, and usualy i don`t care about 3 GB loss. But I`m interested in so-called "deep technical knowledge"   EMC`s white papers do not give a definition what exactly those 3 GB contains. They say just "metadata"

5.7K Posts

September 14th, 2012 02:00

Same here. I did not see this coming.

November 24th, 2014 06:00

VNX 1 use 1GB slices. So that is 1024 slices to to track and store in metadata for a 1TiB LUN.

3% is the overhead for LUN creation.

When I create a 1TiB LUN (1024GiB) I get an overhead of 31GiB. 3.4% on 1024 so pretty close.

Maybe the smallest overhead it can use is 3GiB?

247 Posts

November 25th, 2014 00:00

Correct PacmanSAN, there's always a minimum of 3GB of overhead.

No Events found!

Top