thin and zeroing

Question

Hi,

I am looking a bit into thin, thick (eager and lazy) and trying to understand how VAAI benefits it. Today someone explained to me that regardless of thin or thick, before writing to a block, it has to be zeroed. That means that for example, when using thin, first VMAX allocates the space, then zeroes it, and finally it writes the data. Is this true? If so, why does it do that? Should not it be more efficient just allocating and writing? Eventually it is the same output if it was zeroed or not, the new data will remove what was before (zeroes or anything else). What is the purpose of the zeroing?

Thanks

codyhosterman · Accepted Answer

The reason for zeroing is a security one. VMware wants to assure that a guest OS does not read stale data that might be left on a device from a previous VM or external host, so before any reads or writes to a new sector of the virtual disk the ESXi kernel first writes zeroes (or uses WRITE SAME) then reads or writes the data.

There is more information on using different types of virtual disks with VMAX virtual provisioning in this one paper I wrote and maintain:

http://www.emc.com/collateral/hardware/white-papers/h6813-implting-symmetrix-vrtl-prvsning-vsphere-wp.pdf

codyhosterman · Answer

Another one of mine is just about VAAI which hopefully you find helpful! http://www.emc.com/collateral/hardware/white-papers/h8115-vmware-vstorage-vmax-wp.pdf

zgz_rever · Answer

Thanks! You solved my question.

By the way, I am trying to find if VAAI helps thin somehow. Just talking about WRITE SAME. My conclusion is that the only reason why VAAI can help thin to reduce the performance penalty in comparison with thick is because it helps when zeroing after allocating space.

I just read somewhere that thanks to VAAI, thin and thick have no performance differences anymore, but I think that is true only in very specific cases, if my conclusion is correct.Thanks

codyhosterman · Answer

I wouldnt say no performance differences, but it is greatly lessened as the first write penalty caused by the zeroing on demand nature of thin or zeroedthick virtuals disks is almost gone. There is still some penalty from issuing write same but it is almost negligible. In the second document I have linked there is some performance data on the introduction of WRITE SAME. The first document I linked has a discussion on using thin virtual disks versus thick with the VMAX that goes into more detail. In my opinion performance is not the primary topic for deciding on which type any more--space efficiency or protection against space exhaustion might be more important (or just more likely to make a difference).

Because in reality, most writes (except for the initial creation and install of the OS and applications) are not going to un-initialized regions so there is essentially no difference in performance in those cases.

Zikas · Answer

Hi Cody i totally agreed with you and i would like to add that a difference between thin and thick provisioning from the ESX host is the monitoring!!!!!!

When you "serve" the device inside the ESX as a thin, then the ESX admin has to motitor the space limit in order to avoid any limitation issues but also and the Storage Admin has to monitor the devices.

When you "serve" the device inside the ESX as a thich, then the ESX admin has nothing to do with the monitoring of their ESX's devices.

codyhosterman · Answer

Definitely! Which is why I said that your level of concern over protection against space exhaustion is more important than your concern with performance at this point. Most of my discussion in the white paper is focused on that. For the most part it comes down to balancing requirements with this image:

Zikas · Answer

Thank you very much Cody.

Your answer and your figure is much appreciated.

I mentioned that and actually i tried to point it out that, because many of the ESX admins does not care about the space exhaustion, they are not monitoring the space when they have thin devices and they have on their minds only the performance.

Have a nice evening.

codyhosterman · Answer

Sure thing! Yeah monitoring is the last thing some people think about unfortunately... And thin-on-thin really makes that an important topic to consider like you said--especially who should do the monitoring.

Thanks you have a good evening too! (still in the morning here though )

Zikas · Answer

Dynamox, hello, it's been a long time......... Probably you and your team belongs to a special 'specie' of ESX admin. My ESX admins are a little bit stubborns:)

dynamox · Answer

NY Yankees wrote:Thank you very much Cody.Your answer and your figure is much appreciated.I mentioned that and actually i tried to point it out that, because many of the ESX admins does not care about the space exhaustion, they are not monitoring the space when they have thin devices and they have on their minds only the performance.my ESX admins don't care about performance, they care about capacity. They gladly handed over performance responsibility to me (storage team)

Zikas · Answer

Cody, hairs have grown on my tongue, to tell that to the ESX admins all the time.

Here in Greece the time is 21:25.

You don't know how much i enjoy the community. My knowledge has skyrocket especially in VMWARE matters!!!

Happy to learn from the experts.

Have a day then and i hope that i 'll meet you pretty soon in the Community.

Zikas · Answer

Actually the monitoring is being performed by us. The EMC RESIDENTS!!!!!!!!! Very usefull 'tool' for the customers!!!!!!!!!

Symmetrix

thin and zeroing

Was this post helpful?