On 09/09/14 17:38, Christos Grivas wrote: > Hello all, > > I have a fundamental question which, hopefully, someone here can answer. > I will ask by presenting a probably non-realistic example. > > Lets say I have 10 compute nodes as separate physical machines with 1 > core each. Without taking into account the ability to overload each core > (the famous 16:1) I would like to know: > > *Can i use all the 10 cores to create a single instance that takes > advantage of the all the core pool?* Or the fact that each compute node > has its own hypervisor limits the core usage of each instance to one? > Are the whole resources abstracted and seen as one or not? > > Thank you in advance. Hi Chris, Thanks for looking into OpenStack! Your question has been asked and answered previously at: https://ask.openstack.org/en/question/1230/are-cpu-memory-pooled/ https://ask.openstack.org/en/question/32877/multiple-compute-for-one-instance-is-it-possible/ https://ask.openstack.org/en/question/7925/can-vms-use-paralleled-cpus/ https://ask.openstack.org/en/question/6932/cores-from-multiple-physical-hosts-in-vm/ However, there has been some work going on recently around NUMA - that might be worth looking into if you have such a requirement. Regards, Tom