VMFS block size: bigger is better

[dingding]

the default block size is 1MB, maximum is 8MB. since storage is so cheaper. every time you create VMFS, i suggest choose 8MB, it will increase performance and when you want to extend your VMFS, you needn't to reformat the VMFS because of the size limit. agree with me?

one of my customer increase the big file deployment performance(shorter time) by increase block size.

is there any disadvantage by choosing bigger block size?

i heard of windows VM can increase performance by align partition, is there any method for linux VM? and how to do it? thanks in advance!

[Rajeev_S]

Hi,

I've read that increasing block size would increase in performance. But havent really observered the performance benefit .

You can align the partition in linux using belklow steps,

1. Type n to create a new partition.

2. Type p to create a primary partition.

3. Type 1 to create partition No. 1.

4. Select the defaults to use the complete disk.

5. Type t to set the partition's system ID.

6. Type fb to set the partition system ID to 83 (Linux).

7. Type x to go into expert mode.

8. Type b to adjust the starting block number.

9. Type 1 to choose partition 1.

10. Type 128 to set it to 128 (the array's stripe element size).

11. Type w to write label and partition information to disk.

Hope this helps !

[amvmware]

My understanding is there is no noticeable performance difference from selecting a particular block size. You should choose a block size based on what you intend to do with the VMFS datastore and the maximum file size you will look to create on that datastore.

Don't forget to leave points for helpful/correct posts.

[bluebull2rhyme]

Any improvement in I/O when selecting 8MB than 1MB? Really it's tough situation to find it being not a storage guy

[amvmware]

I have seen it suggested that as vSphere can now grow a VMFS datastore that if you need to increase the size of a vmdk file - using an 8mb block size from the start means you can expand easily - whereas using a 1MB block size to create a 500Gb vmdk file would not be possible and you would have to move the VM or create the vmdk file onto a datatstore that would support that size of vmdk file and the appropriate block size.

I find that it pays to understand your vmfs datastore requirements at the start of a project and provide datastores with different block sizes - this is a typical scenario that customers overlooked in the rush to virtualise and it comes back to bite you.

Don't forget to leave points for helpful/correct posts.