Thin provisioning - how can I see total allocated vs. avail

[kherrmann]

Hey all, we are preparing to embark down the path of using thin-provisioned VVs. I've been digging around trying to find how people manage using thin-provisioning without shooting themselves in the foot. I'm hoping the community has some insights from their experiences.

One question I have is, how do you effectively use CPG growth warnings? If you have a number of CPGs growing evenly, you could effectively use a lot of space without the warnings going off. Of course, we'd put alerts at the system level, for sure, but curious how people decide the appropriate sizing for their warnings.

Another issue that's been concerning me is reporting the storage utilization vs. allocation. Is there an easy way to see the total allocated vs. used? I'd like to be able to quickly compare total storage available in the system, total used, and total allocated, so we can see for each storage tier how close we are to filling up available space, and how overallocated we are.

Beyond that, what metrics do you find really help you get a grip on your TP implementation? I suspect I will learn a lot from actually doing it, but any help to avoid stupid mistakes would be appreciated greatly.

[chunklet]

One thing to keep in mind is that using thin provisioning is that you still only have finite disk resources. If you don't know that you are going to be able to grow your system before you run out of space, you probably would only want to provision thin in limited circumstances. If everything is thin and you run out of space, everything stops. Of course you could set maximum sizes on the CPG's in the background, but then what is it getting you? I personally would not run thin.

What's your OS's?

If youre running windows, set your vols small and grow when needed. If youre running vmware, start with fewer datastores and add more when you need to.

[kherrmann]

We're mainly considering Thin for our vmware use. While we work hard to help our customers understand the flexible nature of VMWare, and how we can fairly easily grow their storage, it is often the case that their stated needs are still far above the actual.

VMWare seems like a great use case for TP because of this very common issue. We are planning on sticking with thick-provisioning at the OS or hypervisor level, and let the 3PAR work its TP magic on the backend.

One thing I'd like to avoid is overprovisioning to any significant degree. I'd like to keep the TP overhead well within the physical limits of the T800 or P10800. What TP then will help with is just making sure that our customers get the storage they feel they need (within reason) while not wasting storage they will never use. So, understanding how much we actually have provisioned vs. used is critical.

Monitoring Virtual Size seems like it is close to what I want, but I'd prefer to see the raw numbers, similar to the graph on the summary screen of the system. All I see there is Failed, Free and Allocated. Does Allocated take into account TP? If so, then it would be useful to also be able to see of all the allocated space, what is actually in use.

[chunklet]

We spent a lot of time comparing raw numbers. If I recall correctly, the sum of space used in User Raw Reserved, Snap Raw Reserved and Admin Raw Reserved = Total Raw used by a volume.

[kherrmann]

Thanks.

In another TP related question...we are planning on using Thick provisioned VMDKs. We were told that using Zero detect, we'd still get a lot of the benefits of thin on thin, but with simplified administration.

And that works great at the beginning of a VMs life, but does anyone know how one can have the OS convert deleted files to zeros? Is there anyway to have that process automated? I'm concerned that over the life of a VMDK it'll just clutter with random non-zero blocks and limit the advantages of TP.

[Josh82]

To convert deleted files inside a vmdk to zeros, you would use sdelete.

If you're running server 2012 on server 2012 vhdx then you don't need to do anything, as the vhdx is UNMAP capable. So the vm would unmap to hyper-v, and hyper-v would unmap to the array.....

[mujzeptu]

V400 here. I had understood it that running our VMFS datastores on 3par TP suite with all vms eager zero'ed would be the best for performance/efficiency?

Are others finding thin on thin to be ideal? Also I was curious on your windows vms if you have sdelete run automatically and if so how often? Do you -Z or -C?

[Cleanur]

Because 3PAR works by growing to meet demand, there are no upfront reservations and as such your free space really depends on which CPG’s are growing and how each CPG has been directed to grow. E.g Raid 10 = 100% to 50% overhead, Raid 5 could be 33% to 11%, Raid 6 could be 33% to 14% overhead.

Since there are no upfront reservations you need to forecast growth based on historical growth rates, which is where system reporter comes in. However if you want to know how much free space you have available to a particular CPG and estimate how much you would have if a CPG were to become fully allocated, you can use “showspace – cpg *” the * is just a wildcard i.e list everything.

Basically the EstFree column shows how much space is free per CPG under RawFree and the LDFree estimates how much usable capacity you could get from this Raw space based on the CPG Raid type. Now divide the LDFree by 1024 for the CPG to provide you with the GB’s of usable space available. This info is also available within the GUI, look at a particular CPG and the "Estimated Free System Space" will match this number.

If you want to look at the system as a whole just look at “initialized free space” per drive class in the system capacity summary to see available free space for all CPG's, this is really the number you need to monitor.

[kherrmann]

One of the things that concerns me greatly about TP is if we get overzealous adding TP VVs. What if we have 25% free space on NL, and a customer asks for 10% of the total disk. Then another customer requests 20%. We know they are unlikely to use half the space they requested, so we are comfortable adding the next customer who requests an additional 10%. So, now, if they all max out the space they were allocated, we are now X% above our total capacity.

This isn't an issue when we still have space to grow in an array, as we would just add more disks, but when we are getting close to that ceiling, it gets scary.

How can we see the total potential for all the VVs we've allocated? That would be a useful stat to see just how overprovisioned we are.

[Richard Siemers]

There is a showspace command that may help. Its useful for calculating freespace based on cpg settings, to include HA and Raid overhead.