Streaming Iceberg Nexmark observations

2,000,000 deterministic events; parallelism 4; four-bucket input/output tables. All times include streaming completion and final drain/commit.

Query	State / mini-batch	Variant	StreamFusion with stock Iceberg IO median [range], s	StreamFusion median [range], s	Median ratio
q3	disk-off	exact	0.629 [0.621–0.752]	0.611 [0.608–0.731]	1.03×
q3	memory-off	exact	0.676 [0.525–0.808]	0.618 [0.582–0.683]	1.09×
q3	memory-on	exact	0.440 [0.430–0.481]	0.422 [0.419–0.449]	1.04×
q4	disk-off	exact	47.419 [19.557–50.285]	17.557 [14.450–37.906]	2.70×
q4	memory-off	exact	9.980 [3.026–14.397]	21.511 [5.434–30.415]	0.46×
q4	memory-on	exact	1.800 [1.595–22.173]	3.521 [1.320–9.591]	0.51×

Exact-query aggregates

Geometric means use every exact-query cell in each mode; labelled expression variants remain separate above. Throughput is nominal corpus events divided by median end-to-end time, including startup, checkpoints and drain.

State / mini-batch	Queries	Geomean ratio	StreamFusion with stock Iceberg IO corpus events/s, range	StreamFusion corpus events/s, range
disk-off	2	1.67×	42,177–3,181,999	113,918–3,274,728
memory-off	2	0.71×	200,400–2,960,139	92,975–3,235,522
memory-on	2	0.73×	1,110,922–4,545,899	567,949–4,743,924

Headline sampling convention

Memory uses the first measured sample; disk uses the best of the first two. The repeated medians and complete ranges above remain the primary results.

Query	State / mini-batch	Variant	StreamFusion with stock Iceberg IO, s	StreamFusion, s	Ratio
q3	disk-off	exact	0.621	0.611	1.02×
q3	memory-off	exact	0.808	0.683	1.18×
q3	memory-on	exact	0.481	0.419	1.15×
q4	disk-off	exact	19.557	14.450	1.35×
q4	memory-off	exact	14.397	30.415	0.47×
q4	memory-on	exact	22.173	3.521	6.30×

Raw samples, warmups, parity checks, plans, committed table locations and native/fallback task/file witnesses are retained in this artifact directory. PROCTIME q12 validates per-bidder count conservation and includes waiting for the final processing-time window.