256 MiB capacity · callback iterations 0 · WriteSession

What does this report measure?

Producer submission throughput and cost: how quickly messages enter the queue and how much producer time each message takes.

Compare TwoBufQueue, ReaderWriterQueue, rigtorp/SPSCQueue and atomic_queue with 8 / 16 / 32 / 64 B messages.TwoBufQueue uses a single-block WriteSession for consecutive writes; the outer loop acquires the next block when the current one fills.Callback iterations are 0; consumers still receive messages and invoke the callback.

Procedure: Prepare messages → time consecutive producer submissions → stop timing → wait for the round to drain. Every queue uses identical message counts and lengths.

Timing scope: Timing includes producer payload copying, progress publication, block acquisition/switching during sending and retry waits. Input preparation, end-of-round handoff and callback draining are outside timing. Concurrent copy and consumer threads still affect actual producer speed.

Throughput: Successful submissions divided by accumulated producer send time, in M messages/s (million messages per second); higher is better. GB/s uses net payload bytes.

p50, p99 and p999 are quantiles of producer message-submission cost; lower is better.To measure short operations accurately, time each group of 1,000 successful submissions and divide its duration by 1,000 to obtain one mean per-message cost sample.

p50 (50th percentile): Within a run, 50% of group samples have a mean per-message cost no greater than this value.

p99 (99th percentile): Within a run, 99% of group samples have a mean per-message cost no greater than this value.

p999 (99.9th percentile): Within a run, 99.9% of group samples have a mean per-message cost no greater than this value.

Compute these quantiles independently per run, then average the three run results. The measurement is Producer message-submission overhead: quantiles use 1,000-write group means, not individual write calls or producer-to-consumer end-to-end latency.

Throughput overview

Four queues are grouped for each message length; throughput is in M messages/s.

Complete-payload and ordering validation passedCommon configuration selected from four metricsOne tuned configuration per queue in each report

Each TwoBufQueue data pool is fixed to 256 MiB, totaling 512 MiB. Native SPSC message storage has a 256 MiB budget, including spare slots and padding; control metadata is counted separately.

Each queue uses one configuration for throughput, p50, p99 and p999, shared across 8 / 16 / 32 / 64 B.Selection weights are 35% / 30% / 20% / 15%, with equal participation of all four message lengths. Native capacities derive from a fixed byte budget; copy strategies retain the fixed payload implementation for each length.

The table now uses three search-stage measurements of the common configuration, so it differs from earlier independent per-metric selections. Original records remain; the additional ReaderWriterQueue neighborhood is saved separately.

8 BThroughput · M messages/s

TwoBufQueue

2,306.55

ReaderWriterQueue

801.00

rigtorp/SPSCQueue

326.11

atomic_queue

186.30

16 BThroughput · M messages/s

TwoBufQueue

1,697.73

ReaderWriterQueue

767.38

rigtorp/SPSCQueue

267.48

atomic_queue

48.65

32 BThroughput · M messages/s

TwoBufQueue

658.60

ReaderWriterQueue

422.63

rigtorp/SPSCQueue

220.68

atomic_queue

30.20

64 BThroughput · M messages/s

TwoBufQueue

383.21

ReaderWriterQueue

204.26

rigtorp/SPSCQueue

99.42

atomic_queue

45.19

Performance comparison

Equal message lengths share a background color; the entire TwoBufQueue row is bold and outlined. Below throughput, the minimum and maximum of three runs are shown.

Message length

Implementation / copy strategy	Throughput M messages/s	Payload throughput GB/s	Producer submission cost p50 ns/message	Producer submission cost p99 ns/message	Producer submission cost p999 ns/message	Backpressure %
8 BCallback iterations 0 · 10,000,000 messages per run · mean of three runs
TwoBufQueueSingle-block WriteSession	2,306.5461,955.353 – 2,687.720	18.452	0.292	1.389	1.833	0%0 / 30,000,000
ReaderWriterQueueComplete member copy	801.002649.604 – 893.901	6.408	1.222	1.972	2.694	0%0 / 30,000,000
rigtorp/SPSCQueueMember copy + destination write prefetch	326.105313.386 – 333.183	2.609	3.000	4.347	6.708	0%0 / 30,000,000
atomic_queueComplete member copy	186.298175.399 – 198.343	1.490	4.986	17.583	36.542	0%0 / 30,000,000
16 BCallback iterations 0 · 10,000,000 messages per run · mean of three runs
TwoBufQueueSingle-block WriteSession	1,697.7261,469.391 – 2,014.335	27.164	0.472	1.653	2.917	0%0 / 30,000,000
ReaderWriterQueueZero destination, then copy the full payload with memcpy	767.375724.706 – 800.675	12.278	1.319	2.569	5.653	0%0 / 30,000,000
rigtorp/SPSCQueueMember copy + destination write prefetch	267.481230.185 – 304.328	4.280	3.097	5.861	8.861	0%0 / 30,000,000
atomic_queueComplete member copy	48.65543.680 – 52.157	0.778	13.375	40.083	43.278	0%0 / 30,000,000
32 BCallback iterations 0 · 10,000,000 messages per run · mean of three runs
TwoBufQueueSingle-block WriteSession	658.601391.339 – 835.602	21.075	1.028	6.222	10.306	0%0 / 30,000,000
ReaderWriterQueueZero destination, then copy the full payload with memcpy	422.627390.659 – 440.068	13.524	2.667	4.903	9.722	0%0 / 30,000,000
rigtorp/SPSCQueueMember copy + destination write prefetch	220.680172.547 – 252.079	7.062	4.056	14.097	17.292	0%0 / 30,000,000
atomic_queueComplete member copy	30.20228.745 – 31.363	0.966	35.847	69.181	314.000	0%0 / 30,000,000
64 BCallback iterations 0 · 10,000,000 messages per run · mean of three runs
TwoBufQueueSingle-block WriteSession	383.213355.437 – 408.296	24.526	2.208	7.625	11.542	0%0 / 30,000,000
ReaderWriterQueueComplete member copy	204.257142.833 – 239.658	13.072	4.681	9.875	20.361	0%0 / 30,000,000
rigtorp/SPSCQueueMember copy + destination write prefetch	99.41565.287 – 126.390	6.363	7.361	22.861	26.194	0%0 / 30,000,000
atomic_queueComplete member copy	45.19339.567 – 48.690	2.892	29.764	40.028	56.222	0%0 / 30,000,000

The table's p50 / p99 / p999 measure Time spent submitting messages in the producer thread: first calculate the mean cost of each group of Mean cost of 1,000 successful writes quantiles, averaged over three runs; these are neither individual-call nor end-to-end latency. Every metric in a row uses one configuration, shared across all four lengths for that queue. Combined selection makes tradeoffs; individual metrics are no longer claimed to be independently optimal. Click any metric to inspect that same complete configuration.Backpressure % = messages affected by backpressure ÷ total messages × 100%, using combined totals from three throughput runs; numerator and denominator are also shown below. A message counts once even if several write attempts fail.

Three individual results

All three measurements are retained, including slow results. All four metrics use the same configuration ID.

8 BThree individual results and their mean

Implementation / metric	Run 1	Run 2	Run 3	Mean
TwoBufQueueThroughput M messages/s · config c16e7206959f	1,955.353	2,276.565	2,687.720	2,306.546
TwoBufQueuep50 ns/message · config c16e7206959f	0.292	0.292	0.292	0.292
TwoBufQueuep99 ns/message · config c16e7206959f	1.333	1.375	1.458	1.389
TwoBufQueuep999 ns/message · config c16e7206959f	1.875	1.833	1.792	1.833
ReaderWriterQueueThroughput M messages/s · config 05c6a693ebc0	859.500	649.604	893.901	801.002
ReaderWriterQueuep50 ns/message · config 05c6a693ebc0	1.208	1.208	1.250	1.222
ReaderWriterQueuep99 ns/message · config 05c6a693ebc0	2.042	1.833	2.042	1.972
ReaderWriterQueuep999 ns/message · config 05c6a693ebc0	3.417	2.125	2.542	2.694
rigtorp/SPSCQueueThroughput M messages/s · config ddbed3127220	331.746	333.183	313.386	326.105
rigtorp/SPSCQueuep50 ns/message · config ddbed3127220	3.125	2.917	2.958	3.000
rigtorp/SPSCQueuep99 ns/message · config ddbed3127220	4.583	4.208	4.250	4.347
rigtorp/SPSCQueuep999 ns/message · config ddbed3127220	7.042	6.583	6.500	6.708
atomic_queueThroughput M messages/s · config f54226336788	198.343	175.399	185.152	186.298
atomic_queuep50 ns/message · config f54226336788	4.917	5.000	5.042	4.986
atomic_queuep99 ns/message · config f54226336788	7.167	8.875	36.708	17.583
atomic_queuep999 ns/message · config f54226336788	37.500	34.708	37.417	36.542

16 BThree individual results and their mean

Implementation / metric	Run 1	Run 2	Run 3	Mean
TwoBufQueueThroughput M messages/s · config d43262717095	1,469.391	1,609.453	2,014.335	1,697.726
TwoBufQueuep50 ns/message · config d43262717095	0.458	0.500	0.458	0.472
TwoBufQueuep99 ns/message · config d43262717095	1.667	1.750	1.542	1.653
TwoBufQueuep999 ns/message · config d43262717095	2.542	3.833	2.375	2.917
ReaderWriterQueueThroughput M messages/s · config 1fe946d9a7bc	724.706	800.675	776.744	767.375
ReaderWriterQueuep50 ns/message · config 1fe946d9a7bc	1.208	1.500	1.250	1.319
ReaderWriterQueuep99 ns/message · config 1fe946d9a7bc	2.375	2.958	2.375	2.569
ReaderWriterQueuep999 ns/message · config 1fe946d9a7bc	3.375	9.542	4.042	5.653
rigtorp/SPSCQueueThroughput M messages/s · config 488489a115af	304.328	230.185	267.931	267.481
rigtorp/SPSCQueuep50 ns/message · config 488489a115af	3.042	3.208	3.042	3.097
rigtorp/SPSCQueuep99 ns/message · config 488489a115af	5.750	6.375	5.458	5.861
rigtorp/SPSCQueuep999 ns/message · config 488489a115af	9.417	9.583	7.583	8.861
atomic_queueThroughput M messages/s · config 7afaa762ae1a	50.128	52.157	43.680	48.655
atomic_queuep50 ns/message · config 7afaa762ae1a	6.625	12.625	20.875	13.375
atomic_queuep99 ns/message · config 7afaa762ae1a	38.375	40.875	41.000	40.083
atomic_queuep999 ns/message · config 7afaa762ae1a	42.292	43.667	43.875	43.278

32 BThree individual results and their mean

Implementation / metric	Run 1	Run 2	Run 3	Mean
TwoBufQueueThroughput M messages/s · config 1b2d5f417aa7	391.339	835.602	748.860	658.601
TwoBufQueuep50 ns/message · config 1b2d5f417aa7	1.000	1.042	1.042	1.028
TwoBufQueuep99 ns/message · config 1b2d5f417aa7	3.958	7.500	7.208	6.222
TwoBufQueuep999 ns/message · config 1b2d5f417aa7	8.458	10.792	11.667	10.306
ReaderWriterQueueThroughput M messages/s · config 37b8e391eac3	437.154	440.068	390.659	422.627
ReaderWriterQueuep50 ns/message · config 37b8e391eac3	3.042	2.958	2.000	2.667
ReaderWriterQueuep99 ns/message · config 37b8e391eac3	6.167	4.208	4.333	4.903
ReaderWriterQueuep999 ns/message · config 37b8e391eac3	10.750	9.500	8.917	9.722
rigtorp/SPSCQueueThroughput M messages/s · config 626a79b8b850	237.414	172.547	252.079	220.680
rigtorp/SPSCQueuep50 ns/message · config 626a79b8b850	4.042	4.042	4.083	4.056
rigtorp/SPSCQueuep99 ns/message · config 626a79b8b850	13.917	15.000	13.375	14.097
rigtorp/SPSCQueuep999 ns/message · config 626a79b8b850	15.833	18.875	17.167	17.292
atomic_queueThroughput M messages/s · config b998250847cc	31.363	28.745	30.499	30.202
atomic_queuep50 ns/message · config b998250847cc	35.042	37.000	35.500	35.847
atomic_queuep99 ns/message · config b998250847cc	110.958	41.833	54.750	69.181
atomic_queuep999 ns/message · config b998250847cc	818.792	54.000	69.208	314.000

64 BThree individual results and their mean

Implementation / metric	Run 1	Run 2	Run 3	Mean
TwoBufQueueThroughput M messages/s · config 2c0b1cb7032e	408.296	355.437	385.908	383.213
TwoBufQueuep50 ns/message · config 2c0b1cb7032e	1.958	2.292	2.375	2.208
TwoBufQueuep99 ns/message · config 2c0b1cb7032e	7.500	7.708	7.667	7.625
TwoBufQueuep999 ns/message · config 2c0b1cb7032e	8.792	12.583	13.250	11.542
ReaderWriterQueueThroughput M messages/s · config 8f16777e2f59	142.833	230.279	239.658	204.257
ReaderWriterQueuep50 ns/message · config 8f16777e2f59	4.875	5.083	4.083	4.681
ReaderWriterQueuep99 ns/message · config 8f16777e2f59	11.125	10.792	7.708	9.875
ReaderWriterQueuep999 ns/message · config 8f16777e2f59	30.708	15.333	15.042	20.361
rigtorp/SPSCQueueThroughput M messages/s · config 0d1082002fb0	65.287	106.570	126.390	99.415
rigtorp/SPSCQueuep50 ns/message · config 0d1082002fb0	7.750	6.875	7.458	7.361
rigtorp/SPSCQueuep99 ns/message · config 0d1082002fb0	22.958	22.250	23.375	22.861
rigtorp/SPSCQueuep999 ns/message · config 0d1082002fb0	26.542	25.125	26.917	26.194
atomic_queueThroughput M messages/s · config b4753ca6ef75	39.567	48.690	47.322	45.193
atomic_queuep50 ns/message · config b4753ca6ef75	34.292	25.375	29.625	29.764
atomic_queuep99 ns/message · config b4753ca6ef75	41.417	38.792	39.875	40.028
atomic_queuep999 ns/message · config b4753ca6ef75	78.833	44.708	45.125	56.222

Throughput and latency each have three runs; modes and random seeds are recorded as run. Each p999 uses approximately 10,000 group samples, leaving about 10 tail samples and sensitivity to scheduling variation.

All initialization parameters

Initialization is listed once per queue and shared by all message lengths. Per-length values such as slot counts use derivation rules; complete run settings are retained in exported JSON and raw records.

TwoBufQueue: initialization shared by all message lengths

All 61 test-class parameters and constructor/template arguments are listed. Message lengths use one shared value list; derived capacities and fixed per-length copy choices use by_payload_bytes mappings instead of four repeated parameter sets.

Test-class parameters

Parameter	Value
transport	two_buf_queue
mode	["throughput", "write_latency"]
payload_bytes	[8, 16, 32, 64]
messages	10000000
warmup_messages	50000
callback_iterations	0
verify_payload	false
prepared_input	true
variable_payload	true
prepared_input_limit_bytes	268435456
measure_callback_time	true
producer_interval_ns	0
source_block_size	33554432
source_pool_min	8
source_pool_max	8
source_queue_capacity	512
writer_backpressure	drop
writer_spin_count	1000
writer_wait_timeout_ms	5000
output_block_size	268435456
output_frame_capacity	1024
output_pool_min	1
output_pool_max	1
output_queue_capacity	512
cancel_claim_mode	tail
output_mode	spsc
output_reader_mode	wait_signal
output_reader_count	1
output_retry_wait_us	1000
output_enable_stats	false
is_scan_current	false
reader_spin_count	1000
rescan_interval_ns	1000000
rescan_max_interval_ns	1000000
consumer_cpu	-1
producer_cpu	-1
producer_qos	user_interactive
round_storage_bytes	214748364
fixed_round_boundaries	true
spsc_buffer_bytes	0
prefault_producer_buffers	true
finish_timeout_ms	5000
work_wait_min_ms	1
finish_retry_ms	1
publish_count	4294967295
use_write_session	true
rigtorp_write_prefetch	true
readerwriterqueue_copy_after_zero	false
spsc_capacity	0
retry_yield_every	1024
spsc_idle_spins	1000
spsc_idle_sleep_ns	0
timer	auto
sample_every	32
max_samples	1000000
latency_batch_size	1000
timer_calibration_samples	16384
call_batch_size	256
random_seed	[42, 43, 44]
timeout_seconds	180
buffer_budget_bytes	536870912

Constructor and template arguments

Parameter	Value
source_thread_id	0
read_layer1_thread_count	1
writer_queue_enable_stats	false
message_prefix_bytes	0
writer_count	1
source_copy_thread_count	1
callback_thread_count	1
pool_depth_instrumentation	false
callback_signature	(thread_id, data, payload_bytes)
callback_business_iterations	0
writer_type	OwnedWriter<Writer<4294967295, false, uint16_t, void>>
writer_api	bounded single-block WriteSession
ordinary_write_readback	false
sync_context	void
callback_wrapper	FixedRecordCallback<PayloadBytes, Sink>
source_pool_payload_bytes	268435456
reader_pool_payload_bytes	268435456

ReaderWriterQueue: initialization shared by all message lengths

Test-class parameters

Parameter	Value
transport	readerwriterqueue
mode	["throughput", "write_latency"]
payload_bytes	[8, 16, 32, 64]
messages	10000000
warmup_messages	50000
callback_iterations	0
verify_payload	false
prepared_input	true
variable_payload	true
prepared_input_limit_bytes	268435456
measure_callback_time	true
producer_interval_ns	0
source_block_size	65536
source_pool_min	4096
source_pool_max	4096
source_queue_capacity	4096
writer_backpressure	drop
writer_spin_count	1000
writer_wait_timeout_ms	5000
output_block_size	1048576
output_frame_capacity	1024
output_pool_min	256
output_pool_max	256
output_queue_capacity	512
cancel_claim_mode	tail
output_mode	spsc
output_reader_mode	wait_signal
output_reader_count	1
output_retry_wait_us	1000
output_enable_stats	false
is_scan_current	false
reader_spin_count	1000
rescan_interval_ns	1000000
rescan_max_interval_ns	1000000
consumer_cpu	-1
producer_cpu	-1
producer_qos	user_interactive
round_storage_bytes	214748364
fixed_round_boundaries	true
spsc_buffer_bytes	268435456
prefault_producer_buffers	true
finish_timeout_ms	5000
work_wait_min_ms	1
finish_retry_ms	1
publish_count	4294967295
use_write_session	false
rigtorp_write_prefetch	false
readerwriterqueue_copy_after_zero	{"by_payload_bytes": {"8": false, "16": true, "32": true, "64": false}}
spsc_capacity	{"by_payload_bytes": {"8": 33488385, "16": 16743937, "32": 8371713, "64": 4185601}}
retry_yield_every	1024
spsc_idle_spins	0
spsc_idle_sleep_ns	12500
timer	auto
sample_every	32
max_samples	1000000
latency_batch_size	1000
timer_calibration_samples	16384
call_batch_size	256
random_seed	[42, 43, 44]
timeout_seconds	180
buffer_budget_bytes	536870912

Constructor and template arguments

Parameter	Value
source_thread_id	0
read_layer1_thread_count	1
writer_queue_enable_stats	false
message_prefix_bytes	0
writer_count	1
source_copy_thread_count	0
callback_thread_count	1
pool_depth_instrumentation	false
callback_signature	(thread_id, data, payload_bytes)
callback_business_iterations	0
unused_fields	source_* / output_* / reader_spin_count / rescan_* / is_scan_current / publish_count: native SPSC does not construct TwoBufQueue layers
upstream_unmodified	true
requested_native_capacity	{"by_payload_bytes": {"8": 33488385, "16": 16743937, "32": 8371713, "64": 4185601}}
native_template	ReaderWriterQueue<Payload,512>
copy_policy	{"by_payload_bytes": {"8": "full member copy", "16": "zero initialization then full memcpy", "32": "zero initialization then full memcpy", "64": "full member copy"}}
consumer_operation	peek → callback → pop

rigtorp/SPSCQueue: initialization shared by all message lengths

Test-class parameters

Parameter	Value
transport	rigtorp
mode	["throughput", "write_latency"]
payload_bytes	[8, 16, 32, 64]
messages	10000000
warmup_messages	50000
callback_iterations	0
verify_payload	false
prepared_input	true
variable_payload	true
prepared_input_limit_bytes	268435456
measure_callback_time	true
producer_interval_ns	0
source_block_size	65536
source_pool_min	4096
source_pool_max	4096
source_queue_capacity	4096
writer_backpressure	drop
writer_spin_count	1000
writer_wait_timeout_ms	5000
output_block_size	1048576
output_frame_capacity	1024
output_pool_min	256
output_pool_max	256
output_queue_capacity	512
cancel_claim_mode	tail
output_mode	spsc
output_reader_mode	wait_signal
output_reader_count	1
output_retry_wait_us	1000
output_enable_stats	false
is_scan_current	false
reader_spin_count	1000
rescan_interval_ns	1000000
rescan_max_interval_ns	1000000
consumer_cpu	-1
producer_cpu	-1
producer_qos	user_interactive
round_storage_bytes	214748364
fixed_round_boundaries	true
spsc_buffer_bytes	268435456
prefault_producer_buffers	true
finish_timeout_ms	5000
work_wait_min_ms	1
finish_retry_ms	1
publish_count	4294967295
use_write_session	false
rigtorp_write_prefetch	true
readerwriterqueue_copy_after_zero	false
spsc_capacity	{"by_payload_bytes": {"8": 33554415, "16": 16777207, "32": 8388603, "64": 4194301}}
retry_yield_every	1024
spsc_idle_spins	4000
spsc_idle_sleep_ns	1000
timer	auto
sample_every	32
max_samples	1000000
latency_batch_size	1000
timer_calibration_samples	16384
call_batch_size	256
random_seed	[42, 43, 44]
timeout_seconds	180
buffer_budget_bytes	536870912

Constructor and template arguments

Parameter	Value
source_thread_id	0
read_layer1_thread_count	1
writer_queue_enable_stats	false
message_prefix_bytes	0
writer_count	1
source_copy_thread_count	0
callback_thread_count	1
pool_depth_instrumentation	false
callback_signature	(thread_id, data, payload_bytes)
callback_business_iterations	0
unused_fields	source_* / output_* / reader_spin_count / rescan_* / is_scan_current / publish_count: native SPSC does not construct TwoBufQueue layers
upstream_unmodified	true
requested_native_capacity	{"by_payload_bytes": {"8": 33554415, "16": 16777207, "32": 8388603, "64": 4194301}}
native_template	rigtorp::SPSCQueue<Payload>
copy_policy	member initialization copy + destination write prefetch
consumer_operation	front → callback → pop

atomic_queue: initialization shared by all message lengths

Test-class parameters

Parameter	Value
transport	atomic_queue
mode	["throughput", "write_latency"]
payload_bytes	[8, 16, 32, 64]
messages	10000000
warmup_messages	50000
callback_iterations	0
verify_payload	false
prepared_input	true
variable_payload	true
prepared_input_limit_bytes	268435456
measure_callback_time	true
producer_interval_ns	0
source_block_size	65536
source_pool_min	4096
source_pool_max	4096
source_queue_capacity	4096
writer_backpressure	drop
writer_spin_count	1000
writer_wait_timeout_ms	5000
output_block_size	1048576
output_frame_capacity	1024
output_pool_min	256
output_pool_max	256
output_queue_capacity	512
cancel_claim_mode	tail
output_mode	spsc
output_reader_mode	wait_signal
output_reader_count	1
output_retry_wait_us	1000
output_enable_stats	false
is_scan_current	false
reader_spin_count	1000
rescan_interval_ns	1000000
rescan_max_interval_ns	1000000
consumer_cpu	-1
producer_cpu	-1
producer_qos	user_interactive
round_storage_bytes	214748364
fixed_round_boundaries	true
spsc_buffer_bytes	268435456
prefault_producer_buffers	true
finish_timeout_ms	5000
work_wait_min_ms	1
finish_retry_ms	1
publish_count	4294967295
use_write_session	false
rigtorp_write_prefetch	false
readerwriterqueue_copy_after_zero	false
spsc_capacity	{"by_payload_bytes": {"8": 33554432, "16": 16777216, "32": 8388608, "64": 4194304}}
retry_yield_every	1024
spsc_idle_spins	2000
spsc_idle_sleep_ns	10000
timer	auto
sample_every	32
max_samples	1000000
latency_batch_size	1000
timer_calibration_samples	16384
call_batch_size	256
random_seed	[42, 43, 44]
timeout_seconds	180
buffer_budget_bytes	536870912

Constructor and template arguments

Parameter	Value
source_thread_id	0
read_layer1_thread_count	1
writer_queue_enable_stats	false
message_prefix_bytes	0
writer_count	1
source_copy_thread_count	0
callback_thread_count	1
pool_depth_instrumentation	false
callback_signature	(thread_id, data, payload_bytes)
callback_business_iterations	0
unused_fields	source_* / output_* / reader_spin_count / rescan_* / is_scan_current / publish_count: native SPSC does not construct TwoBufQueue layers
upstream_unmodified	true
requested_native_capacity	{"by_payload_bytes": {"8": 33554432, "16": 16777216, "32": 8388608, "64": 4194304}}
native_template	AtomicQueueB2<Payload,std::allocator<Payload>,true,false,true>
copy_policy	full member copy
consumer_operation	try_pop → callback

Methodology and environment

Retain the existing results, payload, copy strategy and selection rules.

Fixed test procedure

10 million messages per run, 50,000 warmup messages, callback iterations 0. Input is prepared before write timing, and every payload word varies.

Fixed round limit: 214,748,364 B, equal to 80% of the 256 MiB layer-2 budget. All queues use identical round boundaries; the final round ends at the actual message count.

Only accumulated producer write time is timed. End-of-round handoff, tail publication and callback draining occur outside timing; every message must still be consumed. TwoBufQueue copy and callback threads run independently.

Ordinary write reads back the first byte after the complete copy; Session does not add this readback.

Plateau rules for combined selection

Use common configurations measured at all four lengths. Writer candidates are three neighboring block sizes growing by powers of two; native SPSC candidates are three neighboring spin or wait settings.

The four-length composite scores of three neighboring configurations differ by at most 10%; the coefficient of variation of composite scores grouped by run number is at most 20%. Individual results, variation and neighborhood changes for every length are retained.

Every configuration passes complete validation. Three throughput and three latency runs succeed, with zero write failures, zero Writer backpressure and resolvable latency. Reference coefficients of variation remain capped at 20% / 25% / 40% / 60%.

All metrics use one configuration. Search-stage results are not mixed with earlier independent rechecks. This is the best combined plateau within the measured common configurations, not proof of a global optimum.

Host and build

macOS-14.4.1-arm64-arm-64bit · arm64 · 8 logical CPUs.

Clang 18.1.8 · Release · -O3 -DNDEBUG · no sanitizer.

Producer user_interactive QoS, without hard affinity; copy and callback threads are unpinned. Timer: mach_absolute_time; 1,000 writes per timing group.

Additional measurements use the original executable with identical SHA-256. Upstream headers, complete payloads and copy strategies are unchanged. AC power was confirmed beforehand.

Data and auditability

The original batch contains 1,284 complete validations and 9,708 measurements, with 0 failed runs. The additional ReaderWriterQueue neighborhood contains 24 complete validations and 144 measurements, with 0 failures.

The three reports share the same native SPSC controls. Copy strategies are fixed as disclosed in the table; they and compiler-generated code also affect results.

Earlier per-metric selections are backed up. All four metrics here use common configurations; reference values were fixed before additional measurements. Detailed run settings record actual mode and random_seed values.

Measured executable SHA-256: 1a898890a865e21c4842f8f40f56c3cf9bbfdde912ff2de86782a5c9a3eb0493

Configuration selection record Selection program Additional ReaderWriterQueue measurements Original search records Original candidate list Measurement audit Environment and dependencies Measured source Report generator