VMDK vs RDM: Performance for vSAN

Benoire · Sun Jan 21, 2018 9:20 am

Hi

I'm currently planning my shfit from VMWare vSAN (VMUG Advantage) to starwind vSAN for various reasons. I've currently got access to Perc6i's or H200 on my Dell R710 units and wondered if anyone had undertaken performance tests on the three storage styles within ESXi? I can setup RDM fairly easily as these are SAS HBAs so I wondered how the performance differed and which one users preferred the best, looking for any experience!

Thanks,

Chris

PoSaP · Mon Jan 22, 2018 4:19 pm

Hi!
Had no hands-on experience with exactly the same hardware, but I bet the guys from StarWind would advise following their recommendations when setting up all this stuff.

Benoire · Tue Jan 23, 2018 1:24 am

Yeah I figured they would. Haven't seen any of the blog or 'white paper' releases for this topic so perhaps it might be good if they could do some official testing and release a preference.

Tue Jan 23, 2018 9:32 am

Benoire,

At present we recommend using Thick Provisioned Eager Zeroed VMDKs for ESXi datastores:

Note: Each hard disk should be Thick Provisioned Eager Zeroed.

Source: StarWind Virtual SAN® Hyper-Converged 2 Nodes Scenario with VMware vSphere

Benoire · Fri Jan 26, 2018 1:04 am

Thanks. I quite like MS Storage Spaces Simple as it stripes data across the drives and with 2016, you can rebalance to remove drives when you upgrade them. RDM in to windows would allow this ok, but I'm wondering am I likely to get a performance dis-benefit compared to the VMDK option? Appreciate I'll lose portability of a VMDK but as these are fixed local discs, that is not such an issue.

Wed Jan 31, 2018 12:49 am

Benoire,
Feel free to check the Storage Spaces inside the VM. StarWind will work perfect either on SS or VMDK. I would recommend assigning more Compute resources for VM with Storage Spaces. You can test both scenarios, compare it and we will much appreciate if you will share the results with our community.

Benoire · Wed Jan 31, 2018 12:53 am

Hi Ivan

I will do this. I've got SAS and SATA drives which I will test for both VMDK and MS SS (simple). Will run speed tests on both setups to see which one performs better at a drive level and then in a provisioned VM to see if SS via RDM is more performant than a VMDK in a VMDK (essentially what the VMDK option would be). I'll be utilising a 10g connection for the iSCSI and vMotion network so won't be network limited for replication and MPIO.

Wed Jan 31, 2018 1:05 am

Benoire,
Sounds great! Please keep us posted about the testing.