Question

Hybrid cold/hot data ratio non disk size dependent

  • 28 July 2020
  • 2 replies
  • 224 views

Hello team,

I hope that someone can explain to me why, looking to sizer’s recommandations about hybrid configuations, SSD/HDD ratio is not size dependent but form factor dependent. Basically if you select 2 SSD and 3.5”HDD then a minimum 1:1 ratio is possible (SSDx2 + HDDx2) but if you have have SSDx2 and 2.5” HDD then the minimum ratio is 1:2 (SSDx2 + HDDx4).

Is it because 3.5” HDD SAS or SATA are usually faster than 2.5”? It’s not about size or hardware vendor, recommandations seems to changes according to the disks form factor only


2 replies

Userlevel 4
Badge +5

It is difficult to illustrate without seeing the sizer scenarios. Generally speaking, for SSD/HDD hybrid nodes, the Nutanix rule is always 1:2 SSD to HDD ratio to allow for sufficient I/O to destage from the flash tier. The smallest ROBO nodes are the exception, allowing for 2 SSD and 2 HDD to provide additional flash tier resilience to these remotely deployed nodes, and because their workloads do not commonly generate significant I/O.

There are few exceptions to this rule where a 1:1 SSD to HDD ratio is allowed (Dell XC, Lenovo HX, Fujitsu XF platforms due to the number of drive slots) however there are additional considerations regarding the CPU to accommodate the solution. While such configurations are viable and have their uses they are suboptimal for the majority of scenarios.

I would suggest to reach out to Nutanix Software Engineer for a quality conversation on the subject.

 

Hello Alona,

so there is not a golden rule for SSD/HDD ratio but the general rule is 1:2. I asked because some platforms are certified starting from 1:1, not just in the Robo's scenarios, for example the DX360 LFF. The same platform in the SFF form requires 4 HDD, with any CPU or memory size.

I think you're 100% correct when you point out to I/O because restaging from hot to cold requires fast mechanical disk, usually the certified ones are 10.000 rpm SAS, and 4 disks perform better togheter than 2 because the load can be distributed to every single disk.

I'd be interesting to discuss the matter, how can I reach out a software engineer?

Reply