Distributed workloads

[agelosnm]

Hi guys!

I was wonderring if Proxmox can somehow run at distributed workloads. For example let's say you have a cluster of 3 nodes of 512 GB of memory in total.

Node 1 & 2 have 128 GB and the Node 3 has the rest 256 GB. Would it be possible to have VMs of high memory size to distribute theirs? For example a VM of 120 GB of memory to have 100 GB on node 3 and 20 GB on Node 1.

Would that be possible somehow?

 

[bbgeek17]

No.

---

Blockbridge: Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/prox

 

[0xd149e38e]

I wonder a bit about the application. Anyhow, in HPC world people use workload managers like slurm. You define a task, and the resources required, and the the manager assigns it a machine, or queues it if the machine is busy. Thus I think it can be done, but it is probably not a feature of Proxmox right out if the box.

 

[bbgeek17]

the manager assigns it a machine, or queues it if the machine is busy.

that's correct, it does that ^^^, i.e. assigns a single specialized task/app/process which does some job to a single VM/host. It doesn't magically split a random VM to somehow make it communicate cross disparate physical hosts for memory and cpu synchronization. Its hard enough to synchronize a process across multiple NUMA domains on single physical host - best practice is to constrain it to one NUMA domain to avoid latency.

Other than that, Proxmox is a Hypervisor/Virtual Environment manager - not HPC task coordinator.

---

Blockbridge: Ultra low latency all-NVME shared storage for Proxmox - https://www.blockbridge.com/prox