We have a generic 4 node VMware production cluster running all the traditional production Wintel infrastructure services. A small team here has "R Server for Windows" running on VMs running 4x vCPU and 64GB RAM, and they run their R machine learning code on these VMs. When this happens it takes around 10 minutes to do the analytics. They are in a Resource Pool limiting their VMs to an overall maximum CPU allocation. Now they want to reduce improve scalability of the R Server to get their processing time down to (say) 5 minutes, and are asking for these machines to have 16 or even 32 vCPU. They're advisory is that the application can only scale horizontally. I'm not too keen to be allocating any VMs this level of resource, but wondered if you guys had thoughts on the wisdom of such a configuration?
↧