This post is more than 5 years old
6 Posts
0
1608
May 16th, 2017 04:00
Best setting for RDM on VMware ScaleIO
Hi all!
Need some advice on building ScaleIO 2.0.x on 5 servers with 4 disks each.
These servers will hold some nonprod, but useful activity. So I decided to leave 2 disks on each server for RAID1 (mirror) and create there simple hdd-based VMware datastores. And two other disks from each server assign to ScaleIO.
Which path is more efficient:
1. Make RAID0 from 2 disks, apply RAID controller's read/write cache (BBU present) and assign it as 1 RDM disk to ScaleIO
2. Make two separate RAID0 on each server (single disk), apply RAID controller's read/write cache (BBU present) and assign them as 2 RDM disks to ScaleIO
3. Make RAID0 from 2 disks, no cache at all, assign as 1 RDM disk to ScaleIO
4. Make two separate RAID0 on each server (single disk), no cache at all and assign them as 2 RDM disks to ScaleIO
Maybe I've missed something, please advice.
RHasleton1
73 Posts
0
May 16th, 2017 09:00
ScaleIO best practices would be to follow option 2. That is to create individual RAID0 VDs on each disk and let the raid controller back them with the controller cache. This is what we recommend and see in customer environments.
Hope that helps.
johngmr
6 Posts
0
May 17th, 2017 01:00
RickH, thanks for your input. I guess you're right, it is best way to go. And what do you think about enabling write cache? When some servers do have BBU and other may not?
RHasleton1
73 Posts
1
May 17th, 2017 07:00
If some servers have a BBU and others do not, it would be recommended to only force enable write back cache in a test environment, to see what you should be getting performance wise. It is definitely not recommended to enable this setting in a production environment without BBU as if the RAID card dies in the middle of some writes, you may encounter corruption from the data that was in the raid controller cache that was not yet committed to disk. Data integrity is not guaranteed when doing this.
Good luck!
johngmr
6 Posts
0
May 17th, 2017 22:00
Thanks once again, it will be the best option I guess