RUN 60s-1080p · 2026-10-23T10:35:45Z

60 s 1080p watermark splice benchmark

sim1080 · 1920×1080@30 · 128×32 mark · C wm-splice · 1,800-frame segment · 2 IDRs in one moof · 100 measured reps. This page is the full tagged result set, not a projection.

▶ live players #60s-1080p #1-moof #1800-samples #2-IDR #multi-GOP #probe-PASS #strict-decode-PASS #pixel-verified #SLO-p50-FAIL #SLO-p99-FAIL #RSS-129.8MiB

Verdict: 60 s is technically valid and verified — multi-GOP/multi-IDR in one segment works — but it is not deployable under the current per-segment SLO. 4 s is the only tested shape that passes both p50 (≤30 ms) and p99 (≤150 ms). 6 s is the safe upper bound to measure next.

Median wall

150.51 ms

100 reps · single-moof 60 s

p99 wall

161.36 ms

SLO ceiling 150 ms → fail

CPU mean

146.38 ms

0.146 CPU-s per segment

Peak RSS

129.8 MiB

per in-flight 60 s splice

Throughput

6.60 seg/s

single 2-CPU reference

100-seg batch

1.83 s

8 logical cores, ideal

Live players #video #60s #30s #4s

Native HTML5 H.264 players. The 4 s baseline autoplays muted as a loop; the 60 s player is the verified primary output. Both 60 s containers carry the same 1,800 samples, so one progressive MP4 covers both — the two-moof timing remains in the stats above. The lower pair is a 4× zoom of the 160×64 window around the 128×32 watermark region.

Muted by default · native controls · loops

60 s · single-moofVERIFIED1,800 samples2 IDRs

Primary benchmark output · 150.52 ms median · 161.36 ms p99 · 129.8 MiB peak RSSMP4

30 sVERIFIED900 samples1 IDR

68.00 ms median · 73.01 ms p99MP4

4 s baselineLIVE LOOP120 samples1 IDR

10.25 ms median · 11.12 ms p99 · contract baselineMP4

Watermark region · 4× zoom

160×64 crop at (1712,624), scaled 4× nearest. Left: original, no mark. Right: spliced output, region changed. This run uses the PCM placeholder fill, not the production text-plate library.

Original regionno mark

Source, unmodified regionMP4

Spliced regionregion changedPCM

0 non-region pixel diffs at frames 0 / 1500MP4

Players are recorded outputs, looped for preview. This is not a real-time encoder stream. The 60 s player and both zoom players use the verified single-moof output; the two-moof output is container-equivalent after progressive remux (byte-identical MP4).

Run shape #method #60s #1800f

Source / segment

Source: ffmpeg testsrc2 1920x1080@30, 1800 frames
Resolution / fps: 1920x1080 @ 30 fps
Encoder: patched x264 0.164.x slice-map, keyint 900, min-keyint 900, CRF 28, veryfast, CAVLC, bframes 0, no-deblock, threads 1
Encoded: 1,800 frames · 22.37 MB · 83 s · 2982.41 kb/s · encode wall 83
Sequence: keyint 900 / min-keyint 900 → exact 30 s IDR boundaries; 60 s segment contains two 30 s GOPs (IDRs at 0 and 900).
Zone: MB (108,40) · 8×2 MB / 1728×640 px · first_mb 4908 · C=8 R=2
Segment object: 1 moof / 1 trun · 1,800 samples · 2 IDRs · 22.40 MB input; second variant has 2 moofs × 900 samples.

Benchmark method

Binary: /home/bein/wm/tools/c/wm-splice
Binary SHA-256: a98ec6a8aae99c3563f8c32a…
Reps: 60 s 1-moof: 100; 60 s 2-moof: 50; 30 s: 50; 4 s: 100. 3 warm-ups discarded per case.
Execution: one C wm-splice process per request; full C post-probe; output written; warm page cache; sequential; taskset -c 0,1; RLIMIT_AS=4 GiB
Constraint: taskset -c 0,1 + RLIMIT_AS=4 GiB on a 16-logical-CPU / 65.75 GB host; no other benchmark load. This is a 2-CPU constrained reference, not a 2-vCPU VM.
Timing: wall + wait4 user/sys CPU + peak RSS per child; warm page cache.
Verification: C probe / Python probe, stock ffmpeg strict decode, ffprobe keyframes, pixel diff, SHA-256.

Segment-size comparison #fresh #100-reps #SLO

median bar (teal ≤150 ms, red >150 ms)30 ms p50 ceiling150 ms p99 ceiling

4 s baseline10.25 ms median · 10.07 CPU mean

30 s68.00 ms median · 67.36 CPU mean

60 s · single-moof150.51 ms median · 146.38 CPU mean

60 s · two-moof144.80 ms median · 141.61 CPU mean

Shape	Size / output	Median wall	p95	p99	CPU mean	Peak RSS	SLO	Tags / comment
4 s baseline 120 f · 1 IDR · 1 moof	1.52 MB 1.52 MB input · 1.52 MB output (+0.11%)	10.25 ms	10.73 ms	11.12 ms	10.07 ms	10.9 MiB	PASS p50 / PASS p99	#1080p#4s#baseline#1-IDR#SLO-p50-PASS#SLO-p99-PASS#low-RSS Contract-shaped baseline. 15 zone rotations/min, 1,800 requests/2 h, 10.9 MiB peak RSS. Passes both SLOs.
30 s 900 f · 1 IDR · 1 moof	11.18 MB 11.21 MB input · 11.18 MB output (-0.24%)	68.00 ms	72.06 ms	73.01 ms	67.36 ms	65.8 MiB	FAIL p50 / PASS p99	#1080p#30s#1-IDR#exact-30s-keyint-900#SLO-p50-FAIL#SLO-p99-PASS Exact 30 s requires keyint 900; with the original keyint 120 source use 28 s or 32 s. 2 rotations/min, 240 requests/2 h. p50 fails, p99 passes.
60 s · single-moof 1,800 f · 2 IDR · 1 moof	22.35 MB 22.40 MB input · 22.35 MB output (-0.23%)	150.51 ms	159.07 ms	161.36 ms	146.38 ms	129.8 MiB	FAIL p50 / FAIL p99	#1080p#60s#1-moof#1800-samples#2-IDR#multi-GOP#multi-IDR#SLO-p50-FAIL#SLO-p99-FAIL#RSS-129.8MiB 1,800 samples in one moof/trun, IDRs at sample 0 and 900. Mark verified in the second GOP. 1 rotation/min, 120 requests/2 h. p50 and p99 both fail; 129.8 MiB peak RSS.
60 s · two-moof 1,800 f · 2 IDR · 2 moof	22.34 MB 22.39 MB input · 22.34 MB output (-0.23%)	144.80 ms	152.14 ms	155.04 ms	141.61 ms	130.0 MiB	FAIL p50 / FAIL p99	#1080p#60s#2-moof#2-IDR#multi-fragment#SLO-p50-FAIL#SLO-p99-FAIL#RSS-130.0MiB Two 900-sample fragments. Nearly the same latency as the single-moof shape. 120 requests/2 h. p50 and p99 both fail; 130.0 MiB peak RSS.

Fresh runs: 4 s 100 reps, 30 s 50 reps, 60 s 1-moof 100 reps, 60 s 2-moof 50 reps. The 60 s 1-moof shape is the primary run. All four fresh outputs passed the Python probe, the C wm-probe recheck, and stock ffmpeg strict decode (exit 0, 0 stderr bytes); keyframe/pixel detail is shown for the 60 s case. SLO is per segment: p50 ≤30 ms and p99 ≤150 ms.

60 s run detail #60s #100-reps #stats

Single-moof 60 s

Metric	1 moof	2 moof
n	100	50
Wall min	143.60 ms	139.01 ms
Wall median	150.51 ms	144.80 ms
Wall mean	151.57 ms	145.67 ms
Wall p95	159.07 ms	152.14 ms
Wall p99	161.36 ms	155.04 ms
Wall max	163.24 ms	156.79 ms
Wall stdev	5.12 ms	4.48 ms
Wall CV	3.4%	3.1%
CPU mean	146.38 ms	141.61 ms
CPU p95	149.99 ms	144.90 ms
CPU p99	150.72 ms	147.57 ms
Peak RSS	129.8 MiB / 136.1 MB	130.0 MiB / 136.3 MB
Throughput	6.60 segments/s	6.86 segments/s
100 segments CPU	14.64 CPU-s	14.16 CPU-s
100 on 8 cores	1.83 s	1.77 s
100 on 2 cores	7.32 s	7.08 s
120 on 8 cores	2.20 s	2.12 s
120 on 2 cores	8.78 s	8.50 s
SLO p50	FAIL	FAIL
SLO p99	FAIL	FAIL
C probe recheck	PASS (1800 samples, rc 0)	PASS (1800 samples, rc 0)
Output	22.35 MB (-0.23% vs input)	22.34 MB (-0.23% vs input)
SHA-256	1ccdd36f722379d0d6a25d3d…	170e19b0de49f8f609e5b31b…

Wall-time distribution

100 sequential warm-cache splices of the 60 s single-moof segment. Median 150.5 ms; p99 161.4 ms; SLO ceiling 150 ms.

count per binmedianp99

Verification #probe #decode #pixels

Probe

PASS

Python probe: 1800 samples · --expect-substituted · rc 0
C wm-probe recheck: PASS (1800 samples, rc 0)

Strict decode

exit 0

0 stderr bytes under -err_detect explode+crccheck -xerror

Keyframes

0, 900

IDR samples 0 and 900 → two GOPs in one 60 s segment

Pixel proof

frame 0
non-region diffs: 0
region changed: 3864

frame 1500 · second GOP
non-region diffs: 0
region changed: 3844

spliced · frame 0 · region changed (PCM placeholder)

Crop: 160×64 px at (1712,624), scaled 4× nearest. This benchmark uses the fail-open PCM placeholder fill, not the production FreeType text-plate library. The image proves region substitution and persistence in the second GOP; it is not a text-legibility test.

SHA-256: 1ccdd36f722379d0d6a25d3dc4061da3f1f0dace6cc8d8f58280c56bb3ca795f
1ccdd36f722379d0d6a25d3dc4061da3f1f0dace6cc8d8f58280c56bb3ca795f · 22,346,965 bytes

Capacity math #10GB #CPU-seconds #concurrency

100 concurrent 60 s splices on a 10 GB unit: no. 10 GiB ÷ 129.8 MiB = 78.9 with zero overhead; 10 GB decimal ÷ 136.1 MB = 73.5; at 20% headroom that is 59–63. 100 × 129.8 MiB = 12.7 GiB just for the splices.

100 segments every 140 ms: also no for 60 s. One 60 s splice is ~0.146 CPU-s; 100 segments = 14.6 CPU-s. On 8 logical cores that is ~1.83 s, and you would need ~105 cores to finish in 140 ms. The 4 s shape needs only ~8 cores for the same 100/140 ms target.

Metric	4 s	30 s	60 s single-moof	60 s two-moof
Peak RSS (MiB)	10.9	65.8	129.8	130.0
Max concurrent in 10 GiB (zero overhead)	938	156	79	79
Max concurrent in 10 GB decimal (zero overhead)	873	145	73	73
Max concurrent in 10 GiB at 80% headroom	750	124	63	63
CPU mean (ms/segment)	10.07	67.36	146.38	141.61
100 segments = CPU-s	1.01	6.74	14.64	14.16
100 segments on 8 cores (s)	0.13	0.84	1.83	1.77
100 segments on 2 cores (s)	0.50	3.37	7.32	7.08
120 segments = CPU-s	1.21	8.08	17.57	16.99
120 segments on 8 cores (s)	0.15	1.01	2.20	2.12
Cores needed for 100 segments in 140 ms	7.2	48.1	104.6	101.1
100 segments input I/O (GB)	0.152	1.121	2.240	2.239
I/O rate for 100 in 140 ms (GB/s)	1.08	8.01	16.00	15.99

RSS is peak RSS per in-flight process, not per cached segment. CPU-seconds are single-core seconds; wall time scales as CPU-seconds / usable cores (ideal, no I/O or memory contention). Numbers include the current C post-probe and output write. Shared libc pages make RSS a slight overcount vs PSS.

Requests, startup and ABR granularity #requests #startup #seek

Segment	Shape	Size	Requests / 2 h	Download @10 Mb/s	@25 Mb/s	@50 Mb/s	Seek / ABR step	Tags
4 s baseline	120 f · 1 IDR · 1 moof	1.52 MB	1,800	1.21 s	0.49 s	0.24 s	4 s	#1080p#4s#baseline#1-IDR#SLO-p50-PASS#SLO-p99-PASS#low-RSS
30 s	900 f · 1 IDR · 1 moof	11.21 MB	240	8.97 s	3.59 s	1.79 s	30 s	#1080p#30s#1-IDR#exact-30s-keyint-900#SLO-p50-FAIL#SLO-p99-PASS
60 s · single-moof	1,800 f · 2 IDR · 1 moof	22.40 MB	120	17.92 s	7.17 s	3.58 s	60 s	#1080p#60s#1-moof#1800-samples#2-IDR#multi-GOP#multi-IDR#SLO-p50-FAIL#SLO-p99-FAIL#RSS-129.8MiB
60 s · two-moof	1,800 f · 2 IDR · 2 moof	22.39 MB	120	17.91 s	7.17 s	3.58 s	60 s	#1080p#60s#2-moof#2-IDR#multi-fragment#SLO-p50-FAIL#SLO-p99-FAIL#RSS-130.0MiB

Full-segment download assumption. Progressive players may start before the segment completes, but common fMP4 players wait for the full segment; at 60 s the object is 22.4 MB. 30/60 s therefore trades 7.5–15× fewer requests for 7.5–15× coarser startup, seek and ABR steps.

Anti-crop cost #anti-crop #zone-rotation

60 s single-zone
Zone fixed for 1,800 frames. One static crop of the 128×32 region removes the mark for the full minute. 1 rotation/min.

30 s single-zone
Zone fixed for 900 frames. 2 rotations/min. Better than 60 s, still 7.5× less rotation than 4 s.

4 s baseline
15 zone rotations/min. A static crop only removes one 4 s window per minute. This is the anti-crop shape the schedule was designed for.

The 60 s grouped per-GOP test preserves the 4 s rotation inside one delivery object: 15 sub-splices + concat, 234 ms/60 s, strict decode exit 0. But the current whole-file one-schedule probe cannot validate it, and 234 ms is over the 150 ms p99 ceiling. It is not a drop-in path; it would need per-GOP schedule support in one splicer pass.

All per-iteration results #raw #300-runs

Every measured run, not just aggregates. Warm-up runs are excluded. Raw JSON: stats.json · stats.min.json.

4 s baseline — all 100 measured runs (wall ms, CPU ms, user ms, sys ms, peak RSS KB)

i	wall	cpu	user	sys	RSS KB
0	10.42	10.25	6.15	4.10	10624
1	10.09	9.94	5.96	3.98	10596
2	10.29	10.14	7.10	3.04	10596
3	10.41	10.25	5.59	4.66	10620
4	10.40	10.19	5.09	5.09	10608
5	10.08	9.88	5.93	3.95	10596
6	10.03	9.85	5.91	3.94	10596
7	10.60	9.96	6.97	2.99	10624
8	10.11	9.91	6.94	2.97	10596
9	10.02	9.82	6.87	2.94	10524
10	10.09	9.90	4.95	4.95	10516
11	10.15	9.95	5.97	3.98	10536
12	9.88	9.67	7.52	2.15	10536
13	10.22	10.01	5.46	4.55	10568
14	9.84	9.63	6.42	3.21	10596
15	9.82	9.61	6.41	3.20	10600
16	9.75	9.57	6.70	2.87	10624
17	9.75	9.57	4.78	4.78	10616
18	9.74	9.55	6.68	2.86	10624
19	10.46	10.26	5.13	5.13	10512
20	10.77	10.55	7.67	2.88	10592
21	10.35	10.21	7.14	3.06	10620
22	10.44	10.28	5.61	4.67	10596
23	10.16	10.03	6.02	4.01	10620
24	10.40	10.27	7.19	3.08	10624
25	10.18	10.04	6.03	4.02	10592
26	10.66	10.52	5.26	5.26	10524
27	10.17	10.02	7.01	3.00	10604
28	10.37	10.23	5.11	5.11	10596
29	10.12	9.92	4.96	4.96	10624
30	10.28	9.67	6.44	3.22	10616
31	9.92	9.73	4.86	4.86	10524
32	9.94	9.73	7.79	1.95	10568
33	9.86	9.64	6.75	2.89	10512
34	9.68	9.47	7.58	1.89	10520
35	10.00	9.79	5.88	3.92	10592
36	9.78	9.59	6.71	2.88	10512
37	9.94	9.70	3.88	5.82	10592
38	9.79	9.58	6.39	3.19	10616
39	11.24	11.03	6.01	5.01	10612
40	10.32	10.18	5.09	5.09	10512
41	10.26	10.13	6.08	4.05	10620
42	10.28	10.14	6.45	3.69	10624
43	10.57	10.40	6.24	4.16	10616
44	10.27	10.11	6.07	4.04	10624
45	10.38	10.23	8.18	2.04	10512
46	10.25	10.11	5.06	5.06	10508
47	10.33	10.20	6.12	4.08	10536
48	10.31	10.16	5.08	5.08	10608
49	10.55	10.42	4.17	6.25	10616
50	10.40	10.27	5.60	4.67	10616
51	10.15	9.94	6.96	2.98	10592
52	10.52	10.30	5.15	5.15	10536
53	10.01	9.85	6.56	3.28	10516
54	9.75	9.56	5.31	4.25	10592
55	9.93	9.71	6.47	3.23	10620
56	9.75	9.56	5.73	3.82	10620
57	9.96	9.77	5.86	3.91	10616
58	10.04	9.84	4.92	4.92	10592
59	10.61	10.38	4.72	5.67	10612
60	10.06	9.90	5.94	3.96	10564
61	10.39	10.24	6.14	4.10	10516
62	10.47	10.33	5.63	4.69	10604
63	10.24	10.10	6.06	4.04	10592
64	10.12	9.91	5.95	3.96	10568
65	10.62	10.38	6.61	3.77	10512
66	10.28	10.14	7.09	3.04	10620
67	10.16	9.99	5.99	3.99	10592
68	10.00	9.80	5.88	3.92	10616
69	10.12	9.90	4.95	4.95	11184
70	10.52	10.35	5.17	5.17	10624
71	10.29	10.15	4.06	6.09	10608
72	10.57	10.41	4.73	5.68	10520
73	10.10	9.97	5.98	3.99	10612
74	10.71	10.48	4.76	5.72	10512
75	11.12	10.72	4.87	5.85	10524
76	10.77	10.62	7.73	2.90	10512
77	10.13	10.01	6.00	4.00	10524
78	10.72	10.58	6.73	3.85	10608
79	10.29	10.13	4.05	6.08	10596
80	10.54	10.40	7.28	3.12	10512
81	10.29	10.14	5.07	5.07	10524
82	10.30	10.15	4.06	6.09	10620
83	10.13	10.00	8.00	2.00	10620
84	10.23	10.10	5.05	5.05	10612
85	10.21	10.05	5.02	5.02	10604
86	10.26	10.11	6.07	4.05	10596
87	10.67	10.55	3.84	6.71	10616
88	10.12	10.00	6.00	4.00	10620
89	10.40	10.28	5.14	5.14	10516
90	10.03	9.87	4.94	4.94	10592
91	10.44	10.31	7.50	2.81	10592
92	10.15	10.02	4.01	6.01	10620
93	10.71	10.57	4.80	5.77	10616
94	10.91	10.70	5.84	4.86	10604
95	10.44	10.27	5.13	5.13	10624
96	10.08	9.89	4.95	4.95	10536
97	10.12	9.93	3.97	5.96	10624
98	10.34	10.11	7.08	3.03	10596
99	10.01	9.79	5.87	3.92	10524

30 s — all 50 measured runs (wall ms, CPU ms, user ms, sys ms, peak RSS KB)

i	wall	cpu	user	sys	RSS KB
0	67.23	66.95	41.59	25.36	67276
1	69.49	69.23	42.14	27.09	67380
2	69.26	68.99	43.99	25.00	67384
3	67.97	67.66	44.44	23.23	67380
4	68.03	67.30	37.73	29.57	67388
5	68.40	67.98	40.98	26.99	67292
6	66.90	66.64	38.79	27.85	67288
7	67.92	67.65	43.77	23.88	67360
8	66.70	66.46	44.31	22.15	67380
9	67.00	66.70	46.49	20.21	67376
10	67.11	66.84	42.90	23.94	67380
11	66.32	66.08	40.05	26.03	67356
12	67.41	67.10	39.06	28.04	67360
13	68.28	67.98	38.99	28.99	67360
14	69.24	68.98	41.99	26.99	67284
15	67.31	67.02	40.01	27.01	67316
16	69.37	69.10	38.06	31.05	67276
17	67.17	66.91	39.95	26.96	67284
18	68.69	68.42	37.68	30.74	67308
19	73.18	69.90	42.94	26.96	67324
20	67.62	67.36	42.59	24.76	67384
21	71.89	68.12	43.08	25.04	67356
22	70.62	66.79	43.52	23.28	67312
23	66.88	66.56	43.37	23.20	67384
24	66.35	65.92	43.61	22.31	67384
25	68.25	67.94	43.96	23.98	67320
26	66.83	66.57	42.72	23.84	67336
27	71.39	66.78	40.86	25.91	67236
28	72.08	66.51	41.69	24.82	67316
29	68.53	68.29	44.19	24.10	67316
30	66.52	66.24	41.15	25.09	67276
31	68.21	67.95	37.97	29.98	67380
32	67.38	67.12	38.07	29.05	67380
33	68.04	67.79	45.86	21.93	67284
34	67.46	67.21	38.54	28.66	67236
35	72.83	69.00	42.62	26.38	67320
36	69.30	69.03	43.02	26.01	67380
37	70.51	66.87	40.92	25.95	67360
38	67.86	67.56	40.34	27.23	67236
39	66.53	66.22	43.14	23.07	67324
40	66.24	65.97	39.98	25.99	67216
41	66.06	65.77	38.86	26.90	67280
42	66.20	65.89	42.58	23.32	67312
43	70.81	66.63	44.42	22.21	67384
44	72.05	66.25	36.69	29.56	67320
45	71.65	68.10	37.05	31.05	67284
46	68.08	67.80	38.89	28.91	67376
47	66.99	66.69	41.43	25.26	67224
48	68.15	67.82	44.54	23.28	67364
49	67.69	67.42	43.27	24.15	67364

60 s · single-moof — all 100 measured runs (wall ms, CPU ms, user ms, sys ms, peak RSS KB)

i	wall	cpu	user	sys	RSS KB
0	153.19	144.39	86.24	58.16	132940
1	149.98	143.92	89.95	53.97	132916
2	150.46	144.18	89.11	55.07	132936
3	145.05	144.75	83.43	61.32	132844
4	153.58	148.06	89.03	59.02	132856
5	153.75	145.29	86.98	58.31	132836
6	161.22	150.71	89.42	61.29	132944
7	157.42	146.62	92.39	54.23	132856
8	149.50	149.22	86.13	63.09	132916
9	147.34	147.04	92.02	55.01	132912
10	156.05	147.75	87.85	59.90	132944
11	148.32	148.03	84.02	64.02	132924
12	147.76	147.36	84.63	62.73	132936
13	147.83	147.47	90.29	57.18	132836
14	144.74	144.46	76.24	68.22	132844
15	154.07	144.05	81.59	62.45	132844
16	156.18	144.81	86.48	58.32	132832
17	155.73	144.36	86.82	57.54	132936
18	158.79	148.68	88.40	60.27	132936
19	148.35	148.05	88.03	60.02	132936
20	150.07	144.56	83.32	61.24	132928
21	154.91	143.77	89.24	54.53	132936
22	155.03	143.51	89.69	53.82	132944
23	145.32	144.04	84.02	60.02	132924
24	145.39	145.07	85.04	60.03	132908
25	146.16	145.55	86.73	58.82	132944
26	153.51	149.59	93.37	56.22	132836
27	158.33	147.27	87.57	59.70	132944
28	157.55	147.02	93.65	53.37	132944
29	159.01	146.82	80.45	66.37	132936
30	149.06	148.64	88.38	60.26	132936
31	148.77	148.45	93.28	55.17	132836
32	153.54	149.45	83.69	65.76	132832
33	156.62	145.07	93.05	52.03	132928
34	155.63	144.82	84.06	60.76	132932
35	155.04	144.17	80.65	63.52	132928
36	144.40	144.09	84.05	60.04	132924
37	150.15	144.47	86.28	58.19	132936
38	156.27	145.45	86.86	58.58	132832
39	155.16	145.29	86.77	58.52	132844
40	148.80	148.50	81.28	67.23	132844
41	147.41	147.11	94.07	53.04	132888
42	147.80	147.50	90.69	56.81	132936
43	153.39	145.02	86.01	59.01	132840
44	143.80	143.52	86.71	56.81	132832
45	145.00	144.70	88.82	55.89	132844
46	144.49	144.16	83.09	61.07	132844
47	145.35	145.04	94.03	51.01	132912
48	155.32	145.42	83.24	62.18	132832
49	156.92	145.25	87.75	57.49	132940
50	155.62	144.78	88.86	55.91	132888
51	160.64	150.34	90.20	60.13	132836
52	148.50	148.20	89.12	59.08	132888
53	145.20	144.87	83.92	60.94	132836
54	146.28	145.97	86.98	58.99	132912
55	149.06	148.72	86.84	61.88	132932
56	145.45	145.14	83.08	62.06	132944
57	144.85	144.50	88.31	56.19	132944
58	150.57	144.70	88.82	55.88	132936
59	156.26	144.38	86.23	58.16	132840
60	161.34	147.97	89.59	58.38	132940
61	147.87	147.28	89.17	58.11	132928
62	144.59	144.25	91.16	53.09	132936
63	147.84	147.53	90.71	56.82	132940
64	147.32	147.03	88.02	59.01	132912
65	153.38	148.44	91.27	57.17	132924
66	156.32	144.62	95.09	49.53	132940
67	159.05	147.86	90.30	57.56	132832
68	157.78	147.99	89.39	58.60	132944
69	148.12	147.83	89.89	57.93	132912
70	149.71	144.05	83.45	60.60	132928
71	154.79	144.63	84.37	60.26	132908
72	153.38	143.95	80.97	62.98	132828
73	149.13	143.69	89.43	54.26	132924
74	143.60	143.29	96.20	47.10	132916
75	143.76	143.47	84.69	58.78	132936
76	151.89	144.06	88.04	56.02	132916
77	144.10	143.80	89.88	53.92	132936
78	150.31	149.97	91.98	57.99	132844
79	150.15	149.85	89.91	59.94	132936
80	148.35	148.03	90.02	58.01	132936
81	155.88	148.71	98.15	50.56	132836
82	163.24	151.19	91.11	60.08	132884
83	157.88	147.33	91.20	56.12	132840
84	157.16	147.23	86.55	60.68	132912
85	147.55	147.26	87.56	59.70	132852
86	157.59	147.87	85.92	61.95	132928
87	158.95	147.43	86.66	60.77	132912
88	156.36	150.68	87.82	62.87	132844
89	152.95	148.84	85.91	62.93	132832
90	149.13	148.82	90.89	57.93	132836
91	153.60	150.35	91.21	59.14	132940
92	159.44	148.69	91.81	56.88	132840
93	153.43	147.33	81.18	66.15	132836
94	144.66	144.33	91.21	53.12	132832
95	143.68	143.38	88.62	54.76	132916
96	148.30	147.99	88.99	59.00	132832
97	144.06	143.76	88.85	54.91	132940
98	154.54	149.65	93.78	55.87	132916
99	156.09	143.67	81.81	61.86	132836

60 s · two-moof — all 50 measured runs (wall ms, CPU ms, user ms, sys ms, peak RSS KB)

i	wall	cpu	user	sys	RSS KB
0	148.37	142.76	84.45	58.31	133028
1	140.52	140.24	87.15	53.09	133100
2	150.83	139.59	89.74	49.85	133024
3	150.02	139.77	83.46	56.31	133120
4	145.61	139.30	85.57	53.73	133104
5	156.79	143.51	92.33	51.18	133096
6	142.99	142.68	89.43	53.25	133124
7	147.61	140.11	90.07	50.04	133124
8	150.20	140.82	79.46	61.36	133096
9	145.71	139.19	87.12	52.07	133100
10	151.70	140.08	86.43	53.65	133132
11	146.84	146.53	92.71	53.83	133120
12	143.35	143.03	88.02	55.01	133032
13	139.90	139.62	82.77	56.84	133100
14	144.51	144.20	87.52	56.69	133024
15	146.18	141.86	87.91	53.95	133120
16	149.33	142.54	95.36	47.18	133132
17	149.94	145.48	98.32	47.15	133120
18	151.38	143.23	88.14	55.09	133128
19	142.90	142.58	83.75	58.82	133104
20	148.22	143.29	87.17	56.11	133104
21	149.39	142.34	88.59	53.75	133116
22	153.22	148.56	93.72	54.84	133124
23	151.23	142.83	87.90	54.93	133028
24	142.22	141.92	81.95	59.97	133024
25	140.03	139.71	85.82	53.89	133100
26	148.27	139.96	86.59	53.37	133112
27	139.50	139.18	87.11	52.07	133112
28	143.38	143.07	93.05	50.02	133120
29	143.70	143.39	89.24	54.15	133100
30	140.18	139.87	88.55	51.32	133112
31	149.88	139.98	87.62	52.37	133028
32	144.44	139.06	82.04	57.03	133132
33	152.30	142.05	88.03	54.02	133128
34	142.99	142.67	97.46	45.21	133124
35	139.01	138.66	86.41	52.25	133020
36	144.96	139.12	83.08	56.05	133132
37	149.52	143.06	90.04	53.02	133116
38	151.94	142.60	94.74	47.87	133112
39	142.11	141.78	85.87	55.91	133116
40	141.89	141.58	93.72	47.86	133132
41	142.23	141.90	88.94	52.96	133100
42	142.94	142.61	87.37	55.23	133044
43	144.65	143.16	87.10	56.06	133116
44	139.77	139.40	90.61	48.79	133112
45	147.54	139.47	92.31	47.16	133044
46	144.05	143.69	87.42	56.27	133020
47	139.75	139.37	91.24	48.13	133100
48	139.45	139.14	85.47	53.67	133128
49	140.22	139.90	81.94	57.96	133104

Related measured results #context #pilot

Result	Measurement	Tags / comment
120-min 1080p origin encode	10,037 s (167.3 min) for 215,880 frames · ~21.5 fps	#1080p#encode#4s-source#keyint-120 Part 1: 3,615 s + part 2: 6,422 s. This is the 4 s/keyint-120 source used by the original contract run, not the 60 s source.
12-min 1080p C-splice	2,192 ms total · 12.2 ms / 4 s segment · 180 segments	#1080p#4s#c-splice#180-seg Same C binary; warm cache; includes process spawn and post-probe. An earlier 10-segment sample measured 11.3 ms/seg.
4 s 1080p earlier sample	113 ms / 10 segments = 11.3 ms/seg	#1080p#4s#10-seg Earlier point measurement; the fresh 100-rep baseline is 10.25 ms median on the same binary/constraint.
720p 4 s pilot	10.7 ms / segment	#720p#4s#pilot Earlier measured pilot. Included for scale only; it is not directly comparable to the fresh 1080p run.
60 s pilot	146.8 ms single-moof / 138.7 ms two-moof (3 reps)	#60s#pilot#3-reps Pilot before the 100/50-rep run. The fresh medians are 150.5 / 144.8 ms and are the primary results on this page.
60 s grouped per-GOP test	234 ms / 60 s · 15 sub-splices + concat · strict decode exit 0	#60s#per-GOP-zones#anti-crop-preserved#requires-new-probe#p99-FAIL Keeps the 4 s zone rotation inside a 60 s delivery object, but the current whole-file one-schedule probe cannot validate it; 234 ms/60 s is over p99. Not a drop-in path.

Caveats and limits #honesty #scope

Content is synthetic FFmpeg testsrc2 1920×1080@30. Production content, motion and rate control will change per-frame cost and segment size.
Timing is one C wm-splice process per request, including full C post-probe and output write. An in-process server with buffer pooling may be faster; that is not measured here.
Warm page cache, no other benchmark load, sequential requests. Cold cache, parallel load, and 1,000-concurrent AC-5 are not measured.
Pinned to CPUs 0,1 and RLIMIT_AS=4 GiB on a 16-logical-CPU / 62 GB host. This is a constrained reference path, not a 2-vCPU VM.
The benchmark uses the PCM placeholder fill, not the production FreeType text plate. Visual proof shows region substitution, not text legibility.
Only stock ffmpeg software decode was used. No hardware decoder is available on this box (MediaCodec / VideoToolbox / NVDEC / QSV unverified).
30 s exact needs keyint 900; 60 s one-IDR would need keyint 1800. The original 4 s source uses keyint 120, so 30 s is not IDR-aligned there (28 s or 32 s).
SLO is per segment (p50 ≤30 ms, p99 ≤150 ms). If the SLO were redefined per minute, the conclusion would change.
RSS is peak resident set size; shared pages mean PSS would be slightly lower, not enough to fit 100 × 60 s in 10 GB.
Single-moof 60 s is valid ISO BMFF and decodes in ffmpeg; player/CDN behavior beyond this oracle is not verified.

Artifacts #download #repro

verified 60 s single-moof output verified 60 s two-moof output verified 30 s output verified 4 s output 60 s player MP4 30 s player MP4 4 s player MP4 60 s original MP4 (no mark) mark-region original MP4 mark-region spliced MP4 stats.json (full) stats.min.json REPORT60.txt large-segment bench (prior) pixel proof (both GOPs)

Primary run output: out-60s-1moof.m4s · 22.35 MB · SHA-256 1ccdd36f722379d0d6a25d3dc4061da3f1f0dace6cc8d8f58280c56bb3ca795f. Input: fixed60-1moof.m4s · 22.40 MB · SHA-256 033b45b30c249c1e48abfc58f9e596d926d7a673abecce257158b7d44aee336c.