I have a question relating to the amount of free space to leave on a datastore. I recently upgraded to vCenter and with that came along some datastore usage alarm definitions. I see they were set to 75% and 85% as it relates to warning and error thresholds. This got me to thinking, how much should I really be leaving free on my datastore's. I realize this may be subjective given you can't always predict growth needs for a guest VM nor can you predict memory requirements. My philosophy to this point has been to us up as much as I can as to not "waste" knowing I can simply migrate the VM to a new LUN should the need arise to either expand a vdisk or allocate more memory. I'm willing to accept any risk that I wouldn't have time to do a cold migration to a new LUN should the need arise, I simply don't see any situations where I wouldn't be able to accommodate a cold migration to a datastore with sufficient free space.
So my question: am I flawed in my thinking? Am I missing something that could come back to bite me if I simply fill up my datastore's to make the best use of them? Is there a best practice for free space (I'm assuming the 75% and 85% thresholds aren't a best practice, simply a guideline well more like a arbitrary number VMWare threw in w/o much thought behind it). What are others leaving for free space?
Any assistance anyone could provide would be greatly appreciated!
I'm having the same issue. I upgraded to V4 and had red alerts for all my luns. They average between 60 -100GB free and are allocated on 1TB luns. If I go by the settings that come out of the box I need to leave 250GB free on every lun. 10 luns thats 2.5 TB. We have an XP 12000 disk array and I am sorry but there is no way I am wasting 2.5TB of disk space. The cost is way to high. I need someone to clarify if this is best practice. I understand swap files located on datastores but do I really need to be alerted over 75%. Any advice on this greatly appreciated.
As stated above, 20% free space is a good goal. 90% utilized is where you want to draw the hard line in the sand. "None shall pass"...
Consider using a combination of thin and thick provisioned disks in VM's... I've migrated my boot drives to thick provisioned and kept (for now) the additional volumes/drives as thin. Some of those will be converted to thick if performance is better that way. The only way to know there, is through testing. Using the combination should give you a good balance between performance and LUN utilization.
Consider awarding points for "helpful" and/or "correct" answers.
I have been bitten by the 'Snapshot Delta-Delete' datastore free space constaint.
We had a 450GB datastore for VM Storage and a 74GB Datastore for ISO Images, etc..
Took 2 snapshots of a 'Windows f/s' server VM. Each in about a week of each other.
Then to my lacking, forgot to remove the snapshot in a 'very short period of time'.
A few months later - it hit me - I need to remove that snapshot.
Turns out a user had been copying their Outlook .PST to the server each week, without deleting the old one first or overwriting the previous one. Bear in mind this PST was a few GB's. We noticed this process when the Windows Shared Drive space neared 10GB free. Then cleaned up the .PST files.
Guess what? This created some snapshot delta work....
The 'Server VM' has 2 .vmdk files. Total VM datastore usage: 200GB.
There was near 80GB free (77GB) on this datastore: Datastore2.
Rather than deleting the earliest snapshot and then the later - I chose 'Delete All'
- UGH.. The process used up 'ALL' the datastore space. Now the Snapshot(s) was in an 'orphaned' state.
I was able to move one of the other guest VM .vmdk files (30GB) from Datastore2 to the 'datastore1. Then start my 'Server VM'. Used VMConverter to convert the server VM to a 2nd ESXi 4.0 (LAB) host. Then Remove from Inventory and Delete the ESXi (Production) Server Guest VM. Then VM Convert the ESXi (Lab) Host Server VM Guest back to my ESXi 4.0 production host.
All is good.
These days I am almost figuring that the Datastore Remaining Space/Available Space Best Practice - is a moving target.
Much like a MS product. Take the recommended and Double it.
75% is the recommended. 50% if you can - is the safest path.
I've gotten into the model of looking at the 'provisioned space' numbers instead of the space used. Keep ~20% free from the provisioned space standpoint and you should be safe. Especially when you have both thick and thin VMDK files on the LUN/datastore. Using that model/method, I'm keeping over 50% free space on the LUN's, even though I've provisioned up to that 20% mark... This does mean that I have one more LUN than I might have otherwise (three iSCSI LUN's for VM's to use) but that's ok... I'm also using separate LUN's for ISO files as well as another for templates.
I would suggest thinking of thick and thin vmdk files as all thick when you set your LUNs up. Far better to have an extra LUN, or two, than to have to scramble to free up space so that you can do normal functions (like the above post about snapshots)... Probably not a bad idea to set a calendar reminder for yourself every time you create a snapshot so that you don't let it go weeks/months before realizing you need to delete it...
Consider awarding points for "helpful" and/or "correct" answers.
Talking of monitoring Snapshots and the space they are using I've done a blog posting of different ways you can monitor snapshots and setting up alerts for when they grow larger than a specified size
update ticket :
Is there any documentation with recommandation of 20% free on VMWare site?
On lot of community message or other specialist web site, we can found recommandation of 20% free space on datastore.
But on Vmware website, i'v only found this recommandation : http://kb.vmware.com/selfservice/microsites/search.do?cmd=displayKC&externalId=1003412
"200 Megabyte Capacity: VMware suggests 200 MB as the threshold where ESX system administrators take action to resolve capacity issues. VMFS volumes vary widely in size and it is impossible to define a precise percentage, but the risk of unexpected problems become greater at this point."
It doesn't means VMWare recommand to have 200MB free...
I've actually a 8 nodes cluster with HA, DRS, VCB snapshots and I'm not able to prove requierement should be 20% free...
VMWare answer :
500MB free by default.
+ DRS/VMotion : 4 larger VMs memory consumer (if 4 best vm have 8GB : 32GB)
+ space for snapshot : for third tool or regular activity of snapshot : depend of vm size
... but nothing official with a stamped document ...