Interpreting cache size in vsansizer

[Luca82]

Hi everyone, I've a question for you: I dont understand why in a all-flash configuration, the cache is twice bigger than in the hybrid configuration. I show you and example:

- all-flash

- 100vm general purpose with 100gb storage per vm

- Ftt1 Raid1 no dedup

- the result is 28,34 TB RAW and 11TB Usable and 8,1TB cache

- hybrid

- 100vm general purpose with 100gb storage per vm

- Ftt1 Raid1

- the result is 28,34 TB RAW and 11TB Usable and 1,1TB cache

I don't understand why there is this big difference in cache.

Thank you

[TheBobkin]

Hello Luca82​

Had a play with it there and I cannot for the life of me figure out what it is basing the All-Flash Cache-tier sizing calculations on.

For Hybrid it is consistent and makes sense (10% of usable capacity) but the All-Flash one doesn't even make sense for the node-count and the standard configuration of the ReadyNodes offerings (e.g. 2x400GB or 2x800GB Cache-tier is fairly typical of Dell/Cisco 'standard' models (730XD and C240 M5 respectively).

I even noticed that if you use the buttons at the bottom to increase the capacity (which bizarrely doesn't increase the usable capacity, just the node-count) after a certain number of nodes the cache size almost halves and then seems to scale linearly with a reasonable amount per node (900GB) - so maybe there is some calculation error that only is noted at lower node-counts.

After clicking the + button beside the Capacity wheel to add one node:

I will see about getting in touch with the responsible BU for more information unless someone nearer to that side of the business can do so (e.g. depping).

For now, you should be able to get a basic idea of the ratios and sizing from looking at the BOM of some of the most common ReadyNodes like the ones I noted above and the following:

Flash Cache Sizing for All-Flash Configurations | VMware® vSAN™ Design and Sizing Guide | VMware

Bob

[depping]

I have no idea either Paul why this happens. I can ask one of the guys in the field who owns this to respond if possible.

[vmwph]

Let me make sure I understand the inputs first:

* Opening sizer (create project, or not)

* No changes to cluster settings.

* Hybrid, rest of page at defaults

* Workload - single, general purpose, 100 VMs, 100GB per VM.

I'm seeing 11TB usable capacity, 28.34 Raw, but only 6 servers with 100 VMs. So there must be other inputs pushing you up to 9 nodes that I'm missing.

HY-4 profile

Anyway, for the hybrid it's saying 'Raw cache per node' of 183GB.

Since it's 3 disk groups, that means 3 cache disks of ~61GB. Can't buy disks that small, but that doesn't really matter for this.

Rerunning, same inputs as above but all-flash:

* All-flash, rest of page at defaults

* Workload - single, general purpose, 100 VMs, 100GB per VM, RAID-1, Dedup: 1

Same 11TB usable capacity, same 28.34 TB raw. 6 servers. All good so far.

AF-4 profile.

Raw cache per node: 900GB. 3 disk groups = 300GB per cache disk.

Still seems reasonable.

If I understand the question though, why is all-flash asking for a total of 900GB of cache per host vs. hybrid asking for 183GB?

The sizer uses the profiles to determine minimum disk sizes: vSAN Hardware Quick Reference Guide

I've just checked the underlying calculations the sizer is using and it looks like it should be recommending a single 200GB cache disk for the AF-4 profile, but for some reason someone decided to hard-code a minimum of 300GB. Probably because it's going to be impossible to buy smaller disks so it's being 'helpful' vs. being 'accurate'. It should also be recommending that same 200GB cache disk for the HY-4 profile but is reporting the actual cache required.

I'll go ask the eng team to make the hybrid and all-flash calculations consistent. Either they should both be hard-coding minimum drive sizes, or neither should 😉

If you have different inputs that show different results, please let me know and I'll go validate.