Actually 43 VM's 6.5TB, 20% CPU average use (6X Xeon 5680 CPU) 400 write/iops average during 24hours and 3000 of peak.
The goal is two sites in stretched cluster with sync writes, connected with 2x1Gbs fiber and upgradable to 10Gbs if it is necessary to this project.
Actually we have 6 cpu vmware ent. license, 6 cpu Veeam ent.plus license and 6 CPU Microsoft OEM License that i need to rebuy with the new hardware.
One key point is the Oracle VM, we've 6 cpu standard one license and Oracle want to license all Vsphere cluster, we discussed with Oracle Italy and with two legals...and that's it!
Our partner send me two projects variant:
1) 6 nodes dual socket, 3 per site with 12 Haswell v3 cpu 72 cores in global.
2) 6 nodes single socket, 3 per site with 6 Ivy bridge v2 CPU 60 cores in global.
With the number 1 i need to create and buy a dedicated vsphere cluster to run Oracle across 2 of 6 node, I can also use this second cluster to run others VM. 25-30k€ of license difference then the second option.
With the number 2 i'm pretty ok with all license and it's easy to use/manage with only one vsphere cluster.
The question is, can 6 nodes with 6 "old" cpu run good for all my enviroment with inline dedup and all Nutanix features also in case of DR? Our partner, after worloads analisys tell me YES, you can.
Or is it the first option more expensive but "better" in projection for the next 3-4 years?
Thanks for any answer
Best answer by cbrown
While either will work for you today, if you're looking 3-4 years down the line (including growth) option 1 gives you much more room to grow. Not only do you have more cores, but the Haswell processors preform 20%-30% better per core as well:
Because we are entirely software based, the better performance per core means that the IO througput is improved by a similar amount (giving you extra room to grow on that side as well)
Either one will run the full Nutanix stack with all the features, and with our add-node proceedure you can expand the cluster at any time for additional growth space (though I don't know what that would cause Oracle to do...).