I'm in the process of completing an 8 node ScaleIO cluster (infiniband backend) for a company DR project. I'm considering doing a build thread. Would anyone be interested in seeing the setup process and benchmark data?
ScaleIO Compute/Storage Cluster
- Thread starter Jake Sullivan
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
I have one of these setup already. are you planning on using Windows or Linux as your base OS?
Chris
Chris
I used CentOS 7 for my storage nodes and ESXi 6 (update 1) for my compute nodes. I did a little non-standard config with the SVM setup.
1) the workload is almost exclusively 4k-16k files. and lower latency and speed were paramount.
2) Our first choice was to add SSD to an existing SAN array. The $75K to do that was not approved. the $110K (30K for the license and $80K in servers was approved...) we already had the servers. but it still had to be pulled out of a different project.
3) We are a Windows shop. Anything not Windows is almost exclusively excluded.
4) We have 2 DataCore arrays. 1 is now decomm'ed and the other is scheduled to be decommed next week. After 2 outages that took longer than 12 hours to remedy and one of them was a bluescreen in their software (fixed in latest version). management said no more.
5) S2D was not ready yet. - The Server Team was pushing for this solution.
6) We are not Hyper-Converging yet. We look at Hyper-Converged as a point solution that does not offer real scalability and tiered services.
Chris
2) Our first choice was to add SSD to an existing SAN array. The $75K to do that was not approved. the $110K (30K for the license and $80K in servers was approved...) we already had the servers. but it still had to be pulled out of a different project.
3) We are a Windows shop. Anything not Windows is almost exclusively excluded.
4) We have 2 DataCore arrays. 1 is now decomm'ed and the other is scheduled to be decommed next week. After 2 outages that took longer than 12 hours to remedy and one of them was a bluescreen in their software (fixed in latest version). management said no more.
5) S2D was not ready yet. - The Server Team was pushing for this solution.
6) We are not Hyper-Converging yet. We look at Hyper-Converged as a point solution that does not offer real scalability and tiered services.
Chris
The point of hyper-converge is huge cost saving and simple management but from my brief reading and googling, I don't see the big cost saving. For instance, Nutanix license and blessed hardware cost A LOT!
Word. One of the motivations for going with ScaleIO is that it is open licensing. You can pay for support at your option (per TB).
Ceph is not hyper converged. Ceph plus KVM plus open vswitch will get you there
ScaleIO is not cheap for the enterprise level. it is linear pricing. but spending $500K for 1PB license to get 500 TB of usable... it just seems out of reach for most people. that makes Windows DC licenses cheap (S2D) by comparison. even with the new licensing schema.
HyperConverged is good for a Silo or SMB. there looks to be a cost saving. but that is usually hard to find.
Chris
HyperConverged is good for a Silo or SMB. there looks to be a cost saving. but that is usually hard to find.
Chris