StarWind Software

Hello.

I am planning to deploy a 2-node cluster, hyperconverged scenario. It will run banch of virtual machines.
If i not misunderstanding, for maximum i/o localization, i must have a two HA-device.

Should I place this image files on different physical luns? Or i can use one lun/volume on each server, containing two thick-provisioned image?

Assume, each of my servers have 8 HDDs, 2 independent RAID controllers. Options:
a) I create 2 RADI10 from 4 HDDS on each server, and place a ha-image on each own lun
b) I create a RAID10 from 8 HDDS, formatting it in one big volume, place on that two HA-image.
c) I create a RAID10 from 8 HDDS, splitting it in two volumes, place a ha-image on each own volume

What option will be the best from the point of perfomance?
What option will be the best from the point of resyncing HA-images after disaster?

And, i doubt about redundancy. Actually, with RAID10, we have a double overhead in capacity (ok), and then we add a mirror by ha-device. From 8 disks, we getting only 2-disk capacity. Is there any good options to reduce it?

Hi Sergey Pugachov,
In terms of performance and stability I would recommend you to follow option:

a) I create 2 RADI10 from 4 HDDS on each server and place a ha-image on each own lun

You can find the recommended RAID settings following this link:
https://knowledgebase.starwindsoftware. ... ssd-disks/
As for the capacity and the redundancy, you will get 4 disks capacity for cluster storage, but the configuration will depend on future production needs. It could be RAID5 which is slow or RAID0, which is faster but not redundant at all.

My recommendation having a similar setup (if you haven't already configured your Starwind devices) would be 2x large RAID 10 arrays with 2 HA devices on each array.
Reason being that by utilizing both controllers you speed up any potential rebuilds, but having 1 array on each controller simplifies the management for you in future.

I'd recommend multiple HA devices because even though it's additional administration if you have any issues with a HA device there are less resources dependent on that one HA device. I also find that it helps in case of full synchronizations as the HA images will be smaller and therefore sync faster, reducing the time you are without storage redundancy when performing full syncs.

Hope this helps!

cragile,

This is exactly the scenario that would work in most cases.

I supposed, that i should to seek such design: But if I rightly understand Cragile, such option will be more acceptable? I think there should be 2 HA device anyway fro two-node cluster, or we can not achieve IO localization on each node. All the problems, how to store HA-images physically?

Hi Sergey Pugachov,
Actually, RAID 10 from 8 HDDs can show better results than 2 RAID 10 from 4 HDDs. And you can create 2 HA devices on this RAID.
But both ways are acceptable, you can create 2 RAID 10 from 4 HDDs and then put HA device on each of them.