Growing VM disks
-
I was thinking to myself about the growth of file systems on a VMWare system. I hadn't considered it before and was wondering how the growth is handled; is it linear? It couldn't be, right? If I create 3 VMs, and expand two of them to add more space to an existing vmdk, it's tacked on from free space on a file system held by the hypervisor, but then does that mean we're creating a file system on a pre-existing file system, and do the allocation tables themselves generate excess waste? How efficient is that?
#ShowerThoughts -
VMFS is the file system and each VM is a single file on it. Each block has a pointer to it in most filesystems, the only inefficiency in non-linear growth is in the movement of a drive head - which often happens anyway in shared systems regardless of file linearity, and is of no consequence in SSDs. So in modern systems, the idea that fragmentation is going to cause performance issues really isn't an issue. And the complexity of the underlying filesystem doesn't change, because it's just one file per VM anyway.
-
@scottalanmiller said in Growing VM disks:
VMFS is the file system and each VM is a single file on it. Each block has a pointer to it in most filesystems, the only inefficiency in non-linear growth is in the movement of a drive head - which often happens anyway in shared systems regardless of file linearity, and is of no consequence in SSDs. So in modern systems, the idea that fragmentation is going to cause performance issues really isn't an issue. And the complexity of the underlying filesystem doesn't change, because it's just one file per VM anyway.
Unless you add drives to the guest, which gives you multiple files per guest, and then those "drives" or the OS can get grown, which comes back to the allocation table thought, and fragmentation of the file(s). At what point is there a diminishing return?
-
@Grey said in Growing VM disks:
At what point is there a diminishing return?
There never is one. Unlimited fragmentation is assumed in modern storage systems. The overhead of it, if there is any (generally there is none) is already absorbed.
-
@scottalanmiller said in Growing VM disks:
@Grey said in Growing VM disks:
At what point is there a diminishing return?
There never is one. Unlimited fragmentation is assumed in modern storage systems. The overhead of it, if there is any (generally there is none) is already absorbed.
Has anyone done a study?
-
Paging @StorageNinja for insight on this one...
-
So each VM is NOT a single file in VMFS. Each VMDK file is really just a virtual disk characteristic file. The data disk is a -flat.vmdk file (even though we usually don't think of having both of these files when we say VMDK).
-
Maybe these will also help (a bit older though):
https://kb.vmware.com/s/article/1006810 (speaks to VMFS 3 and 5 but not 6)
-
Well, it's crucial to keep them warm and evenly moist until they germinate, then give them plenty of sunshine until they're big enough to transplant...
-
This post is deleted!
