[vlovich123]

You’re right, I did my math with 1 core assumption. Still, it’s 800ms of 100% utilization of all 4 vCPUs across the entire cluster. That’s 40% capacity if the 2s latency was purely CPU bound and the client was next to the server, but we know the initial phase involves some round trips to communicate that 50mib so the 40% is a floor. My point remains that this is several orders of magnitude off in terms of cost still (as you mention network bandwidth required is off by similar orders of magnitude as cpu).