New disk what format choice?

[ikotb]

We add Dell MD3000i SAN in ESX add storage wizard screen....to format the new disk what choices to choose? 256GB block size1MB, 512 2MB, 1024 4MB and 2048 8MB?

Thanks

[mike_laspina]

I always use the maximum, if you ever need to extend the volume you would want the larger block size which max's out at 2TB on 8MB.

[RParker]

We add Dell MD3000i SAN in ESX add storage wizard screen....to format the new disk what choices to choose? 256GB block size1MB, 512 2MB, 1024 4MB and 2048 8MB?

Choose for the LARGEST vmdk file you think you will need because to change this later means you remove the files, format at and copy files back. So whatever you think you will need, that's the block size. 256Gb means you can't have a single file larger than 256Gb on the VMFS volume.

[penghaug]

Hi,

I would say it depends on the size of the LUN you're creating. The lager disk's you want to add a VM the larger LUN and block size you probably want to create.

If you got a 500GB LUN there is no point in creating it 4MB or 8 MB block size.

[mcowger · Accepted solution]

Same here - we always choose the largest just have the flexibility at the loss of a small amount of slack space.

--Matt

[weinstein5]

It depends on how large a virtual disk you want to provide to yur virtual machines - the block size will cap the size of the virtual disk file - 1 MB - 256 GB, 2 MB - 512 GB, 4 MB - 1024 GB, 8 MB - 2048 GB

If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful

[RParker]

Same here - we always choose the largest just have the flexibility at the loss of a small amount of slack space.

I have asked the question before, and not received a sensible answer yet. If it's simply selecting the largest block size, then why do they make this an option? The answer is simple, there MUST be some performance considerations in selecting a larger block size rather than a small one, or something other than simple space constraints (i.e. a small file occupies 8 meg vs 1 meg on a file system, etc..).

That cannot be the ONLY factor, there has to be more to it, otherwise the options wouldn't be in place. This needs to be more research than simply telling everyone to use the largest block size available, have we identified with VM Ware that block size ONLY impact is file space? I don't think so.... There are ALWAYS pro and con for each and every setting, so using 8meg block size MUST have a counterpoint, which is why we don't do it unless we need to.

[mike_laspina]

Hey Parker,

Until now I never checked it out for myself, I have read multiple posts indicating that there was no real difference between 1M and up. So I decided to do a very crude dd test and verify all the bla bla bla.

It turns out that on a local VMFS disk volume there is a significant difference.

root@vh0 root# time dd of=/dev/null if=/vmfs/volumes/vh0-local0/blksize8m

131072+0 records in

131072+0 records out

real 0m5.588s

user 0m0.040s

sys 0m3.340s

root@vh0 root# time dd of=/dev/null if=/vmfs/volumes/vh0-local1/blksize1m

131072+0 records in

131072+0 records out

real 0m4.081s

user 0m0.000s

sys 0m3.090s

The input file is exactly the same on both dd operations and the output file is on identical physical disks models on the same controller and host.

I ran it with at least 15 different combinations the 1M block size was always about 15% faster.

Lesson learned, don't always trust what you read.

Thanks for encouraging me to look.

Regards,

Mike

Just an update - the result seems to be equal perfromance over an FC SAN with the dd command on both 1M and 8M, there does not appear to be a significant difference on a virtualized LUN at this point.

Most likely a disk cache shortage the local system disk test.

Message was edited by: mike.laspina

[BUGCHK]

> (i.e. a small file occupies 8 meg vs 1 meg on a file system

VMFS-3 supports suballocation - multiple small files can share a VMFS block.