Random, uncached reads on the same dataset as the 90/10 - 1.6M/sec, limited by SSD throughput. Cached reads (from the same dataset, with a different distributi

Random, uncached reads on the same dataset as the 90/10 - 1.6M/sec, limited by SSD throughput. Cached reads (from the same dataset, with a different distribution of read keys) you can see on our website at 3.2 M/sec. I should mention that the test clients also run on the same cluster, and in that workload they are a significant fraction of CPU. Read requests come to a storage server with a specific version number attached, so the storage server can reply authoritatively without any further communications. Starting a transaction requires selecting a version number for the read snapshot, and that incurs some latency to deal with the concerns you mention (but scales fine).