> 100% in-memory workload. Don’t even think about hitting disk for 0.0001% of these requests.

They also discuss how these are 1M read requests.