Unsolved
This post is more than 5 years old
2 Posts
0
583
ScaleIO - Multi SDS Node vs Single SDS Node
After getting acquainted with ScaleIO in a Virtual Test Lab we are looking to setup a production environment.
Which would perform better and what at the Pro & Cons of the following typologies?
3x Physical Nodes with 4x Multi SDS's on each
4x SSDs with PCIE Raid Controller Standalone Mode
40Gb Infiniband Dual Ports Adapters
VS
12x Physical Nodes with a Single SDS
1x SSD in each Node with onboard SATA 600 controller
10Gb Dual Port Ethernet Adapter
So what are the Pro's & Co's in regards to Protection Domains, Fault Sets and anything else we have no doubt overlooked?
If performance is similar then it will come down to Cost. Look forward to your input and thanks in advance.
Cheers!
pawelw1
306 Posts
0
November 27th, 2016 04:00
Hi,
If money is not the problem, it would be probably better to go with 12 single SDS. Pros would be:
1. Better resiliency - in 12 nodes environment, when a single machine fails, the impact is very small. In 3 nodes environment, you lose immediately 1/3 of your whole cluster.
2. I can't give you any exact figures, but I'd say performance might be indeed better with 12 nodes - more CPU cycles on each node, more NICs etc.
3. With 12 nodes you can divide them easier into FaultSets (if you plan to use them).
My $0.02 - hope that will help with your considerations :-)
Cheers,
Pawel
Anonymous
5 Practitioner
5 Practitioner
•
274.2K Posts
0
November 27th, 2016 14:00
Hi,
just to add on top of Pawel's answer -
If you install multi SDS nodes - it makes sense to put all of the 4 SDSs in the same node in the same fault set.
Because we don't hold two copies of the some data in the same node, for each Write (for example) 2/3 of your nodes will participate.
whereas, if you create 12 nodes - on each Write - 1/6 of you nodes will participate.
I hope this gives some additional insight
Thanks,
ItaiR
ZMTZIN
2 Posts
0
November 28th, 2016 15:00
Hi,
Appreciate the input all, really helpful information.
We will look to go with 12 Nodes on Commodity Hardware (Dell Optiplex etc) with Dual Port 10Gb Ethernet Adapters.. However the caveat is we need to confirm that the PCIE slots, SATA controller and power wattage requirements are sufficient to obtain maximum performance as this will keep costs down.
Lastly we may deploy hyper-converged as Virtualization Compute Nodes on 8 Threaded Intel CPU's and 32GB Memory with a couple of spare units ready to swap out if a Node goes down.
So lets see how it all goes. Cheers all.
Thanks,
Zeek