Stress Test Results

Long-running stress tests comparing sustained performance between Dekaf and Confluent.Kafka under real-world load.

Last Updated: 2026-08-27 11:20 UTC

info

The paired Dekaf vs Confluent comparison runs weekly (Sunday 2 AM UTC) and updates this page. Manual dispatches stay Dekaf-only unless full_run explicitly requests the same paired publish path. Tests measure sustained performance over 15+ minutes with real Kafka instances.

At a glance

Each row is a like-for-like comparison: both clients run the same sustained workload sequentially on the same VM, and repeated samples are aggregated with a geometric mean across both run orders.

Paired same-VM stress runSustained throughputBroker-confirmed messages per second for the same workload.

DekafConfluentHigher is better

Produce — fire-and-forget

Dekaf1.37M msg/s (1.3×)

Confluent1.01M msg/s

Produce — fire-and-forget (3 brokers)

Dekaf1.03M msg/s (1.6×)

Confluent652.01K msg/s

Produce — acks=all

Dekaf1.19M msg/s (1.3×)

Confluent901.33K msg/s

Produce — acks=all (3 brokers)

Dekaf987.49K msg/s (1.3×)

Confluent738.15K msg/s

Produce — fire-and-forget, idempotent

Dekaf1.25M msg/s (1.3×)

Confluent971.12K msg/s

Produce — fire-and-forget, idempotent (3 brokers)

Dekaf759.04K msg/s (1.2×)

Confluent627.18K msg/s

Produce + consume round-trip

Dekaf3.04M msg/s (2.4×)

Confluent1.25M msg/s

Produce — transactional (exactly-once) (3 brokers)

Dekaf1.10K msg/s (6.9×)

Confluent159 msg/s

Consume — messages

Dekaf1.61M msg/s (1.4×)

Confluent1.13M msg/s

Bars are scaled within each scenario for a direct client-to-client comparison.

Median client CPU timeCPU cost per messageCPU time needed to deliver one message; shorter bars are better.

DekafConfluentLower is better

Produce — fire-and-forget

Dekaf0.82 μs/msg (2.1× less)

Confluent1.75 μs/msg

Produce — fire-and-forget (3 brokers)

Dekaf1.21 μs/msg (1.9× less)

Confluent2.29 μs/msg

Produce — acks=all

Dekaf0.89 μs/msg (2.1× less)

Confluent1.91 μs/msg

Produce — acks=all (3 brokers)

Dekaf1.18 μs/msg (1.8× less)

Confluent2.12 μs/msg

Produce — fire-and-forget, idempotent

Dekaf1.00 μs/msg (1.8× less)

Confluent1.76 μs/msg

Produce — fire-and-forget, idempotent (3 brokers)

Dekaf1.57 μs/msg (1.7× less)

Confluent2.62 μs/msg

Produce + consume round-trip

Dekaf0.98 μs/msg (2.2× less)

Confluent2.17 μs/msg

Produce — transactional (exactly-once) (3 brokers)

Dekaf253.88 μs/msg (1.4× less)

Confluent356.45 μs/msg

Consume — messages

Dekaf0.83 μs/msg (1.6× less)

Confluent1.28 μs/msg

Bars are scaled within each scenario for a direct client-to-client comparison.

Scenario	Dekaf	Confluent	Throughput	CPU per message
Produce — fire-and-forget	1,369,359 msg/s	1,014,593 msg/s	1.3× faster	2.1× less
Produce — fire-and-forget (3 brokers)	1,029,417 msg/s	652,006 msg/s	1.6× faster	1.9× less
Produce — acks=all	1,191,980 msg/s	901,326 msg/s	1.3× faster	2.1× less
Produce — acks=all (3 brokers)	987,486 msg/s	738,149 msg/s	1.3× faster	1.8× less
Produce — fire-and-forget, idempotent	1,245,247 msg/s	971,117 msg/s	1.3× faster	1.8× less
Produce — fire-and-forget, idempotent (3 brokers)	759,039 msg/s	627,177 msg/s	1.2× faster	1.7× less
Produce + consume round-trip	3,037,259 msg/s	1,254,153 msg/s	2.4× faster	2.2× less
Produce — transactional (exactly-once) (3 brokers)	1,095 msg/s	159 msg/s	6.9× faster	1.4× less
Consume — messages	1,611,723 msg/s	1,128,564 msg/s	1.4× faster	1.6× less
Consume — batches	1,654,431 msg/s	—	—	—
Consume — raw bytes	3,020,532 msg/s	—	—	—
Consume — raw byte batches	3,912,585 msg/s	—	—	—

"On par" means within ±5% — differences that small are run-to-run noise. "CPU per message" compares the client CPU cost of delivering one message; "less" means Dekaf needs less CPU. Rows showing "—" have no Confluent counterpart in this run (for example, batch and raw consume APIs that librdkafka does not expose). The full per-run data is below.

Full results

Each section holds the measured per-run data behind the summary: repeated same-VM samples in both client orders, CPU per message and per request, and throughput drift across the run.

Producer (Fire-and-Forget) (15 minutes, 1000B messages)

Order-Balanced Aggregate

Client	Samples	Geomean comparison msg/s	Sample range	Median CPU μs/msg	Comparison Ratio
Dekaf	2	1,369,359	1,292,733–1,450,527	0.82	1.35x
Confluent	2	1,014,593	975,031–1,055,761	1.75	1.00x

The aggregate uses the geometric mean across balanced same-VM samples run in both dekaf-first and confluent-first order. Raw ordered samples remain below.

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf (3conn)	0.76	723.20	1,787,945	1,787,044	-5.7%	-0.31%	1705.12	1,787,945	0	1.35
Dekaf (dekaf-first)	0.74	760.23	1,413,907	1,450,527	-6.1%	-0.50%	1348.41	1,413,907	0	1.04
Dekaf (confluent-first)	0.90	921.54	1,271,562	1,292,733	+10.5%	+0.74%	1212.66	1,271,562	0	1.14
Confluent (confluent-first)	1.74	-	981,375	1,055,761	-26.0%	-2.88%	935.91	981,375	0	1.71
Confluent (dekaf-first)	1.77	-	961,996	975,031	+3.3%	+0.58%	917.43	961,996	0	1.70

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Messages/sec counts broker-confirmed deliveries (end-offset delta). Accepted msg/s is the client-side append rate — a large gap means messages were buffered or dropped without ever reaching the broker.

Producer (Fire-and-Forget), 3 Brokers (15 minutes, 1000B messages)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf (3conn)	1.02	952.18	1,260,341	1,220,523	-20.1%	-1.72%	1201.95	1,260,341	0	1.28
Dekaf	1.21	1114.31	1,040,180	1,029,417	+10.6%	+1.14%	991.99	1,040,180	0	1.26
Confluent	2.29	-	660,092	652,006	-9.2%	-0.83%	629.51	660,092	0	1.51

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Messages/sec counts broker-confirmed deliveries (end-offset delta). Accepted msg/s is the client-side append rate — a large gap means messages were buffered or dropped without ever reaching the broker.

Producer (Acks All) (15 minutes, 1000B messages)

Order-Balanced Aggregate

Client	Samples	Geomean comparison msg/s	Sample range	Median CPU μs/msg	Comparison Ratio
Dekaf	2	1,191,980	1,144,856–1,241,042	0.89	1.32x
Confluent	2	901,326	826,049–983,464	1.91	1.00x

The aggregate uses the geometric mean across balanced same-VM samples run in both dekaf-first and confluent-first order. Raw ordered samples remain below.

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf (dekaf-first)	0.80	766.05	1,230,003	1,241,042	-6.2%	-0.60%	1173.02	1,230,003	0	0.99
Dekaf (confluent-first)	0.98	1001.95	1,140,319	1,144,856	+6.3%	+0.36%	1087.49	1,140,319	0	1.11
Confluent (dekaf-first)	1.75	-	981,544	983,464	-20.7%	-1.56%	936.07	981,544	0	1.72
Confluent (confluent-first)	2.07	-	837,255	826,049	+1.9%	+0.16%	798.47	837,255	0	1.73

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Messages/sec counts broker-confirmed deliveries (end-offset delta). Accepted msg/s is the client-side append rate — a large gap means messages were buffered or dropped without ever reaching the broker.

Producer (Acks All), 3 Brokers (15 minutes, 1000B messages)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf	1.18	1140.16	984,732	987,486	-1.4%	-0.09%	939.11	984,732	0	1.16
Confluent	2.12	-	741,759	738,149	-0.4%	+0.08%	707.40	741,759	0	1.58

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Messages/sec counts broker-confirmed deliveries (end-offset delta). Accepted msg/s is the client-side append rate — a large gap means messages were buffered or dropped without ever reaching the broker.

Producer (Fire-and-Forget, Idempotent) (15 minutes, 1000B messages)

Order-Balanced Aggregate

Client	Samples	Geomean comparison msg/s	Sample range	Median CPU μs/msg	Comparison Ratio
Dekaf	2	1,245,247	1,184,729–1,308,857	1.00	1.28x
Confluent	2	971,117	970,002–972,233	1.76	1.00x

The aggregate uses the geometric mean across balanced same-VM samples run in both dekaf-first and confluent-first order. Raw ordered samples remain below.

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf (3conn)	0.84	787.33	1,645,256	1,593,636	-0.1%	-0.01%	1569.04	1,645,256	0	1.38
Dekaf (confluent-first)	0.91	912.87	1,298,166	1,308,857	+0.5%	-0.04%	1238.03	1,298,166	0	1.18
Dekaf (dekaf-first)	1.10	1124.15	1,174,934	1,184,729	-7.4%	-0.78%	1120.50	1,174,934	0	1.29
Confluent (dekaf-first)	1.75	-	982,564	972,233	+3.9%	+0.63%	937.05	982,564	0	1.72
Confluent (confluent-first)	1.78	-	942,558	970,002	+15.6%	+1.37%	898.89	942,558	0	1.67

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Messages/sec counts broker-confirmed deliveries (end-offset delta). Accepted msg/s is the client-side append rate — a large gap means messages were buffered or dropped without ever reaching the broker.

Producer (Fire-and-Forget, Idempotent), 3 Brokers (15 minutes, 1000B messages)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf (3conn)	1.46	1374.93	842,558	823,261	-16.8%	-1.52%	803.53	842,558	0	1.23
Dekaf	1.57	1554.09	760,858	759,039	+8.3%	+0.65%	725.61	760,858	0	1.20
Confluent	2.62	-	624,978	627,177	+10.5%	+0.86%	596.03	624,978	0	1.63

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Messages/sec counts broker-confirmed deliveries (end-offset delta). Accepted msg/s is the client-side append rate — a large gap means messages were buffered or dropped without ever reaching the broker.

Producer → Consumer Round-Trip Steady State (15 minutes, 128B messages)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf	0.98	5274.00	1,336,226	3,037,259	+44.2%	+689.79%	163.11	1,336,226	0	1.31
Confluent	2.17	-	125,299	1,254,153	+24.0%	+148.48%	15.30	125,299	0	0.27

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Messages/sec counts broker-confirmed deliveries (end-offset delta). Accepted msg/s is the client-side append rate — a large gap means messages were buffered or dropped without ever reaching the broker.

Round-Trip Validation

Client	Expected	Consumed	Missing	Duplicates	Corrupt	Out of Order	Wrong Partition	Unexpected	Timed Out	Result
Confluent	19,792,477	19,792,477	0	0	0	0	0	0	no	PASS
Dekaf	19,792,477	19,792,477	0	0	0	0	0	0	no	PASS

Producer (Transactional EOS), 3 Brokers (15 minutes, 1000B messages)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf	253.88	253.88	808	1,095	+11.5%	+1.09%	0.77	1,078	0	0.27
Confluent	356.45	-	119	159	+3.9%	+0.31%	0.11	158	0	0.06

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Messages/sec counts broker-confirmed deliveries (end-offset delta). Accepted msg/s is the client-side append rate — a large gap means messages were buffered or dropped without ever reaching the broker.

Transaction Verification

Client	Accepted	Committed	Aborted	Delivered	Duplicates	Shortfall	Aborted leaks	Unexpected	Missing sentinels	Status
Confluent	142,300	106,800	35,500	106,800	0	0	0	0	0	PASS
Dekaf	970,000	727,500	242,500	727,500	0	0	0	0	0	PASS

Consumer (15 minutes, 1000B messages, 16,384B seed batches)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf	0.83	-	1,602,152	1,611,723	+8.1%	+0.64%	1527.93	-	0	1.33
Confluent	1.28	-	1,089,952	1,128,564	-5.5%	+0.02%	1039.46	-	0	1.40

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Consumer (Batch) (15 minutes, 1000B messages, 16,384B seed batches)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf	0.80	-	1,652,254	1,654,431	+0.1%	+0.01%	1575.71	-	0	1.33

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Consumer (Raw Bytes) (15 minutes, 1000B messages, 16,384B seed batches)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf	0.52	-	3,057,821	3,020,532	+7.7%	+0.70%	2916.17	-	0	1.58

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Consumer (Raw Batch) (15 minutes, 1000B messages, 16,384B seed batches)

Client	CPU μs/msg	CPU μs/request	Messages/sec	Median msg/s	Drift	Slope %/min	MB/sec	Accepted msg/s	Errors	Standing cores
Dekaf	0.40	-	3,886,412	3,912,585	+11.4%	+1.09%	3706.37	-	0	1.56

Median msg/s is the median sampled client-side throughput interval; it shows steady-state throughput without letting a short late-run stall dominate the whole-run average.

Rows and Comparison Ratio use Median msg/s when available; older result files without interval samples fall back to Messages/sec.

Drift compares last-third with first-third average throughput. Slope is the normalized least-squares trend; steady-state below 85% of peak or slope below -1%/min fails the regression gate.

Memory & GC statistics — latest run

Client	Scenario	Gen0	Gen1	Gen2	Total Allocated	Alloc/msg
Confluent	Consumer	19267	0	0	2229.16 GB	2.38 KB
Confluent	Producer (Fire-and-Forget)	203753	1	1	1059.96 GB	1.26 KB
Confluent	Producer (Fire-and-Forget)	211321	8	1	1039.04 GB	1.26 KB
Confluent	Producer (Fire-and-Forget), 3 Brokers	144044	1	1	712.95 GB	1.26 KB
Confluent	Producer (Acks All)	164320	1	1	904.24 GB	1.26 KB
Confluent	Producer (Acks All)	218254	8	1	1060.16 GB	1.26 KB
Confluent	Producer (Acks All), 3 Brokers	153462	0	0	801.11 GB	1.26 KB
Confluent	Producer (Fire-and-Forget, Idempotent)	187967	1	1	1018.01 GB	1.26 KB
Confluent	Producer (Fire-and-Forget, Idempotent)	218741	48	1	1061.41 GB	1.26 KB
Confluent	Producer (Fire-and-Forget, Idempotent), 3 Brokers	137637	0	0	674.99 GB	1.26 KB
Confluent	Producer → Consumer Round-Trip Steady State	5360	4	4	15.77 GB	856 B
Confluent	Producer (Transactional EOS), 3 Brokers	81	1	1	213.09 MB	1.53 KB
Dekaf	Consumer	23966	15	2	2718.54 GB	1.98 KB
Dekaf	Consumer (Batch)	24703	3	2	2803.87 GB	1.98 KB
Dekaf	Consumer (Raw Bytes)	4	2	1	408.28 MB	0 B
Dekaf	Consumer (Raw Batch)	8	2	1	913.06 MB	0 B
Dekaf	Producer (Fire-and-Forget)	243	3	2	108.84 MB	0 B
Dekaf	Producer (Fire-and-Forget)	217	3	2	839.40 MB	1 B
Dekaf	Producer (Fire-and-Forget), 3 Brokers	126	3	2	235.54 MB	0 B
Dekaf	Producer (Acks All)	171	3	2	130.63 MB	0 B
Dekaf	Producer (Acks All)	178	3	2	654.24 MB	1 B
Dekaf	Producer (Acks All), 3 Brokers	129	3	2	144.51 MB	0 B
Dekaf	Producer (Fire-and-Forget, Idempotent)	177	2	1	115.13 MB	0 B
Dekaf	Producer (Fire-and-Forget, Idempotent)	176	3	1	598.43 MB	1 B
Dekaf	Producer (Fire-and-Forget, Idempotent), 3 Brokers	100	3	2	198.31 MB	0 B
Dekaf	Producer → Consumer Round-Trip Steady State	1101	2	1	286.91 MB	15 B
Dekaf	Producer (Transactional EOS), 3 Brokers	79	1	1	110.58 MB	120 B
Dekaf (3conn)	Producer (Fire-and-Forget)	257	3	2	948.77 MB	1 B
Dekaf (3conn)	Producer (Fire-and-Forget), 3 Brokers	174	4	2	714.18 MB	1 B
Dekaf (3conn)	Producer (Fire-and-Forget, Idempotent)	271	6	2	880.94 MB	1 B
Dekaf (3conn)	Producer (Fire-and-Forget, Idempotent), 3 Brokers	116	3	1	556.14 MB	1 B

Confluent.Kafka uses native librdkafka; .NET GC allocation counters exclude unmanaged allocations.

---

About These Tests

Stress tests measure sustained performance over extended periods against real Kafka brokers, with both clients paired on the same VM for a fair comparison.

Methodology — how these numbers are produced

• Real Kafka: Tests run against actual Apache Kafka instances
• CPU Isolation: Brokers are pinned to dedicated cores and the client under test to its own cores, so the client — not the broker — is the measured bottleneck
• RAM-backed Broker Logs: Kafka log dirs are mounted on tmpfs so disk I/O never caps broker ingestion
• Delivered Throughput: producer tables report broker-confirmed throughput, measured as the end-offset delta across all partitions — not the client-side append rate, which can run far ahead of what the broker ever accepts
• Median Interval Throughput: table order and comparison ratios use median sampled client-side msg/s when available, which is less sensitive to short late-run stalls than the whole-run mean
• Same-VM Pairing: comparable Dekaf and Confluent scenarios run sequentially inside one job/VM; 1-broker producer acceptance lanes run twice in opposite client orders and publish a geometric-mean aggregate, while other lanes alternate order by workflow run number
• Backpressure Parity: both producers are bounded to the same 512 MB local buffer (Dekaf BufferMemory, librdkafka queue.buffering.max) and block on a full buffer, so neither client can absorb an unbounded backlog into RAM
• Consumer Loop Replay: Consumer tests re-read a pre-seeded topic (seek to beginning when drained) instead of racing a live feeder, so the consumer itself is measured; table headings report the 16KB seed batch size because it amplifies per-batch costs relative to well-batched workloads
• Delivery Latency Sampling: 1 in 1000 produced messages is awaited end-to-end to record true broker round-trip latency
• Round-Trip Correctness: Bounded sequenced payloads are consumed back and checked for corruption, wrong partitions, gaps, duplicates, and reordering
• Round-Trip CPU Scope: CPU time covers both bulk production and consumer validation; it is not a producer-only metric
• Round-Trip Alloc Scope: the GC/alloc window likewise spans production plus consume-side validation; values are deliberately consumed as byte[] on both clients for parity, so each consumed payload is materialized as a fresh array (~152 B at 128 B messages) — the expected allocation floor for this lane, not a leak
• CPU Efficiency: CPU time per message differentiates client efficiency even at equal throughput
• Noise-Aware Trends: each scenario is compared with its last 10 matching runs using a median ± 2×MAD band; one adverse excursion warns and two consecutive regressions fail the workflow
• Parallel Execution: Each scenario runs in its own isolated environment
• Both Clients: Direct comparison between Dekaf and Confluent.Kafka
• Memory Monitoring: Tracks GC behavior and memory usage over time
• Error Rates: Ensures stability under load