{ "_note": "Raw per-run output behind docs/efficiency.md. Produced by benchmarks/nexmark_compare/cloud/split-run.sh; not hand-edited. One row per measured run: one engine, one query, one trial, on the same engine node against the same isolated broker, with Flink re-baselined on this provision.", "measured_utc_date": "2026-07-27", "clink_commit": "764e570", "flink_version": "2.2.0-scala_2.12-java21 (upstream image, unmodified)", "supersedes": "efficiency-2026-07-26.json (clink 1eb6fae); this run follows a round of engine fixes", "rig": { "engine_node": "Hetzner ccx23, 4 dedicated vCPU, AMD EPYC Milan, fsn1", "broker_node": "Hetzner ccx23, 4 dedicated vCPU, fsn1, runs the broker alone", "note": "one engine at a time on the engine node; fresh cluster per measured run" }, "workload": { "source": "Kafka topic nx-bid, 4 partitions, 9.2M nexmark bid records", "queries": "nexmark q0 (stateless projection), q12 (windowed GROUP BY)", "parallelism": 4, "sink": "blackhole (output discarded)" }, "how_read": { "cpu": "cgroup v2 cpu.stat usage_usec across every container of the engine under test", "memory": "cgroup v2 memory.stat anon (engine heap+stacks, excludes page cache)", "rate": "the engine's own records-processed counter at 100ms; sustained_slope is the steepest 0.5s slope" }, "correctness_gate": { "harness": "benchmarks/nexmark_compare/run.sh (separate run)", "q0_output_rows": { "clink": 460000, "flink": 460000, "identical": true }, "q12_output_rows": { "clink": 184767, "flink": 184767, "identical": true }, "note": "gate last run at 1eb6fae; the changes since are covered by 1631 core + 835 SQL tests" }, "q12_memory_decomposition": { "groups_195710_mb": 831, "groups_5_mb": 444, "per_group_bytes": 1977, "previous_per_group_bytes": 3914, "note": "Holding records and window fixed and varying only the group count separates per-group state from the group-independent pipeline-buffering floor." }, "not_measured": [ "wall power, PUE, grid carbon intensity, embodied carbon", "workloads beyond nexmark q0 and q12", "hardware other than AMD EPYC Milan", "a tuning effort on either engine", "long-run (days/weeks) memory behaviour" ], "caveat_throughput_variance": "The throughput ratio moves tens of percent between provisions of the same machine type. events_per_cpu_sec and anon_mb hold steady; quote throughput as 'ahead', not to two decimals.", "derived_ratios": { "q0": { "clink_events_per_cpu_sec": 983957, "flink_events_per_cpu_sec": 254425, "cpu_efficiency_ratio": 3.87, "clink_anon_mb": 75.5, "flink_anon_mb": 1169.3, "clink_cores": 0.84, "flink_cores": 2.57, "memory_ratio": 15.5 }, "q12": { "clink_events_per_cpu_sec": 482940, "flink_events_per_cpu_sec": 168993, "cpu_efficiency_ratio": 2.86, "clink_anon_mb": 748.7, "flink_anon_mb": 1625.2, "clink_cores": 1.71, "flink_cores": 2.94, "memory_ratio": 2.2 } }, "runs": [ { "query": "q0", "engine": "clink", "trial": 1, "par": 4, "input_events": 9200000, "sustained_slope": 3978541.8, "drain_rate": 3792293.3, "drain_seconds": 2.426, "reached_target": true, "cpu_seconds": 9.35, "wall_seconds": 11.11, "cores": 0.84, "events_per_cpu_sec": 983957, "anon_mb": 75.5, "peak_mb": 90.6, "mem_fresh_stack": true }, { "query": "q0", "engine": "clink", "trial": 2, "par": 4, "input_events": 9200000, "sustained_slope": 3946703.5, "drain_rate": 3602553.0, "drain_seconds": 2.554, "reached_target": true, "cpu_seconds": 9.53, "wall_seconds": 10.23, "cores": 0.93, "events_per_cpu_sec": 965373, "anon_mb": 76.7, "peak_mb": 100.6, "mem_fresh_stack": true }, { "query": "q0", "engine": "flink", "trial": 1, "par": 4, "input_events": 9200000, "sustained_slope": 2735413.5, "drain_rate": 1156941.6, "drain_seconds": 7.952, "reached_target": true, "cpu_seconds": 36.16, "wall_seconds": 13.86, "cores": 2.61, "events_per_cpu_sec": 254425, "anon_mb": 1176.1, "peak_mb": 1526.7, "mem_fresh_stack": true }, { "query": "q0", "engine": "flink", "trial": 2, "par": 4, "input_events": 9200000, "sustained_slope": 2434241.6, "drain_rate": 1099295.0, "drain_seconds": 8.369, "reached_target": true, "cpu_seconds": 36.51, "wall_seconds": 14.19, "cores": 2.57, "events_per_cpu_sec": 251986, "anon_mb": 1169.3, "peak_mb": 1539.5, "mem_fresh_stack": true }, { "query": "q12", "engine": "clink", "trial": 1, "par": 4, "input_events": 9200000, "sustained_slope": 2904627.5, "drain_rate": 2533107.8, "drain_seconds": 3.632, "reached_target": true, "cpu_seconds": 19.05, "wall_seconds": 11.15, "cores": 1.71, "events_per_cpu_sec": 482940, "anon_mb": 748.7, "peak_mb": 777.0, "mem_fresh_stack": true }, { "query": "q12", "engine": "clink", "trial": 2, "par": 4, "input_events": 9200000, "sustained_slope": 2707807.0, "drain_rate": 1872910.9, "drain_seconds": 4.912, "reached_target": true, "cpu_seconds": 22.6, "wall_seconds": 12.45, "cores": 1.82, "events_per_cpu_sec": 407080, "anon_mb": 889.3, "peak_mb": 917.5, "mem_fresh_stack": true }, { "query": "q12", "engine": "flink", "trial": 1, "par": 4, "input_events": 9200000, "sustained_slope": 1563509.4, "drain_rate": 718693.9, "drain_seconds": 12.801, "reached_target": true, "cpu_seconds": 56.87, "wall_seconds": 19.35, "cores": 2.94, "events_per_cpu_sec": 161772, "anon_mb": 1625.2, "peak_mb": 1863.2, "mem_fresh_stack": true }, { "query": "q12", "engine": "flink", "trial": 2, "par": 4, "input_events": 9200000, "sustained_slope": 1609556.0, "drain_rate": 740263.9, "drain_seconds": 12.428, "reached_target": true, "cpu_seconds": 54.44, "wall_seconds": 18.47, "cores": 2.95, "events_per_cpu_sec": 168993, "anon_mb": 1625.9, "peak_mb": 1860.6, "mem_fresh_stack": true } ] }