Due to the performance hit while zeroing mentioned in the Thin Provisioning Performance white paper this article in the VMware knowledge base could be of some good use.
I would suggest using eagerzeroedthick for any high IO tier 1 type of Virtual Machine. This can be done when creating the VMDK from the GUI by selecting the “Support Clustering Features such as Fault Tolerance” check box.
So go out and check your VMDK’s.
Hey Jon,
Great suggestion. Your link to the white paper isn’t working. I’d be interested to read the reasoning behind the FT performance increase for high IO VMs.
oops! Link to the white paper is working now.
Hey Jon,
Great suggestion. Your link to the white paper isn’t working. I’d be interested to read the reasoning behind the FT performance increase for high IO VMs.
oops! Link to the white paper is working now.
Jon,
I agree that eagerzerothick is great for high I/O VMs, it should be carefully considered as systems that Thin Provision like NFS or V-Max (and many others) will allocate all space once the vmdk is eagerzero.
zerothick continues to play nice with these technologies, but you do pay some I/O penalty for the thin provisioning which continues to be less and less with each VMFS revision.
In our testing, we saw marginal benefits (1-4% performance) from eagerzero over zero running SQL loads. We do have a very large EMC V-Max SAN however to cover some of this I/O. We now only use eagerzero for MSCS VMs.
Good post, and small option many people miss or just overlook!
It is very important to consider especially if you are depending on thin provisioning on the SAN. I do believe some Sans like 3par recover the 0’s but I need to double check.
I have never visibly seen too bad of a performance hit using just zerothick. I always would rather plan ahead to avoid follow up calls ( I do professional service installs) related to slowness.
Another note: It will be slow to deploy a big eagerzeroedthick disk if it is very big.
Jon,
I agree that eagerzerothick is great for high I/O VMs, it should be carefully considered as systems that Thin Provision like NFS or V-Max (and many others) will allocate all space once the vmdk is eagerzero.
zerothick continues to play nice with these technologies, but you do pay some I/O penalty for the thin provisioning which continues to be less and less with each VMFS revision.
In our testing, we saw marginal benefits (1-4% performance) from eagerzero over zero running SQL loads. We do have a very large EMC V-Max SAN however to cover some of this I/O. We now only use eagerzero for MSCS VMs.
Good post, and small option many people miss or just overlook!
It is very important to consider especially if you are depending on thin provisioning on the SAN. I do believe some Sans like 3par recover the 0’s but I need to double check.
I have never visibly seen too bad of a performance hit using just zerothick. I always would rather plan ahead to avoid follow up calls ( I do professional service installs) related to slowness.
Another note: It will be slow to deploy a big eagerzeroedthick disk if it is very big.
Yeah, that comment about recovering the zeroes was my imagination. Storage vendors: start working on that now. You can send the royalties straight to me.
UPDATE: 3par InForm OS 2.3.1 supports Zero detection and can reclaim the 0’s to keep disks thin. This is done by a special ASIC therefore keeping performance impact at near nothing.
Yeah, that comment about recovering the zeroes was my imagination. Storage vendors: start working on that now. You can send the royalties straight to me.
UPDATE: 3par InForm OS 2.3.1 supports Zero detection and can reclaim the 0’s to keep disks thin. This is done by a special ASIC therefore keeping performance impact at near nothing.