Repository navigation
Expand file tree
/
Copy pathdata-pipeline-memory.html
More file actions
1055 lines (1012 loc) · 40.4 KB
/
Copy pathdata-pipeline-memory.html
File metadata and controls
1055 lines (1012 loc) · 40.4 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Waymo Perception Studio — Data Pipeline Memory Map</title>
<style>
* { margin: 0; padding: 0; box-sizing: border-box; }
body {
font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
background: #0a0a0f;
color: #e0e0e8;
padding: 32px;
line-height: 1.5;
}
h1 { font-size: 24px; font-weight: 700; margin-bottom: 4px; color: #fff; }
.subtitle { color: #888; font-size: 14px; margin-bottom: 32px; }
.legend {
display: flex; gap: 20px; flex-wrap: wrap;
margin-bottom: 24px; font-size: 12px;
}
.legend-item { display: flex; align-items: center; gap: 6px; }
.legend-dot {
width: 12px; height: 12px; border-radius: 3px;
}
/* Pipeline Container */
.pipeline {
position: relative;
width: 100%;
overflow-x: auto;
}
/* Phase sections */
.phase {
margin-bottom: 40px;
border-left: 3px solid #333;
padding-left: 20px;
position: relative;
}
.phase::before {
content: '';
position: absolute;
left: -7px; top: 0;
width: 11px; height: 11px;
border-radius: 50%;
background: #555;
border: 2px solid #0a0a0f;
}
.phase.active::before { background: #4ecdc4; }
.phase-header {
display: flex; align-items: baseline; gap: 12px;
margin-bottom: 12px;
}
.phase-num {
font-size: 11px; font-weight: 700;
color: #4ecdc4; text-transform: uppercase;
letter-spacing: 1.5px;
}
.phase-title {
font-size: 16px; font-weight: 600; color: #fff;
}
.phase-time {
font-size: 12px; color: #666;
margin-left: auto;
}
/* Parallel swim lanes */
.swim-lanes {
display: flex;
flex-direction: column;
gap: 6px;
}
.swim-lane {
display: flex;
align-items: center;
gap: 0;
min-height: 44px;
}
.lane-label {
width: 180px; min-width: 180px;
font-size: 11px; color: #999;
text-align: right; padding-right: 12px;
font-weight: 500;
}
.lane-bar-container {
flex: 1;
position: relative;
height: 36px;
}
.lane-bar {
position: absolute;
height: 36px;
border-radius: 6px;
display: flex;
align-items: center;
padding: 0 10px;
font-size: 11px;
font-weight: 600;
white-space: nowrap;
color: rgba(255,255,255,0.9);
transition: opacity 0.2s;
cursor: default;
overflow: hidden;
}
.lane-bar:hover {
opacity: 0.85;
z-index: 10;
}
.lane-bar .mem-tag {
position: absolute;
right: 8px;
font-size: 10px;
font-weight: 700;
background: rgba(0,0,0,0.4);
padding: 2px 6px;
border-radius: 3px;
}
/* Colors */
.c-fetch { background: linear-gradient(135deg, #2d5aa0, #1e3d6f); }
.c-decomp { background: linear-gradient(135deg, #8b5cf6, #6d28d9); }
.c-parse { background: linear-gradient(135deg, #f59e0b, #d97706); }
.c-convert { background: linear-gradient(135deg, #ef4444, #dc2626); }
.c-cache { background: linear-gradient(135deg, #10b981, #059669); }
.c-worker { background: linear-gradient(135deg, #06b6d4, #0891b2); }
.c-gpu { background: linear-gradient(135deg, #f472b6, #ec4899); }
/* Memory chart */
.memory-chart {
margin-top: 40px;
background: #111118;
border-radius: 12px;
padding: 24px;
border: 1px solid #222;
}
.chart-title {
font-size: 15px; font-weight: 600; color: #fff;
margin-bottom: 16px;
}
.chart-svg {
width: 100%; height: 380px;
}
/* Memory table */
.mem-table {
margin-top: 24px;
width: 100%;
border-collapse: collapse;
font-size: 13px;
}
.mem-table th {
text-align: left; padding: 8px 12px;
color: #888; font-weight: 600;
border-bottom: 1px solid #333;
font-size: 11px; text-transform: uppercase;
letter-spacing: 1px;
}
.mem-table td {
padding: 8px 12px;
border-bottom: 1px solid #1a1a24;
}
.mem-table tr:hover td { background: #14141e; }
.mem-val { font-weight: 700; font-variant-numeric: tabular-nums; }
.mem-peak { color: #ef4444; }
.mem-retained { color: #10b981; }
.thread-tag {
display: inline-block;
font-size: 10px; font-weight: 700;
padding: 2px 6px; border-radius: 3px;
text-transform: uppercase; letter-spacing: 0.5px;
}
.thread-main { background: #1e3a5f; color: #60a5fa; }
.thread-worker { background: #3b1764; color: #a78bfa; }
.thread-cam { background: #164e3a; color: #34d399; }
/* Parallel indicator */
.parallel-bracket {
border: 2px dashed #4ecdc4;
border-radius: 8px;
padding: 8px;
margin: 4px 0;
position: relative;
}
.parallel-label {
position: absolute;
top: -10px; left: 12px;
background: #0a0a0f;
padding: 0 6px;
font-size: 10px;
color: #4ecdc4;
font-weight: 700;
text-transform: uppercase;
letter-spacing: 1px;
}
.note {
background: #14141e;
border-left: 3px solid #4ecdc4;
padding: 10px 14px;
font-size: 12px;
color: #aaa;
margin: 8px 0;
border-radius: 0 6px 6px 0;
}
.note strong { color: #fff; }
/* Summary cards */
.summary-cards {
display: grid;
grid-template-columns: repeat(auto-fit, minmax(200px, 1fr));
gap: 16px;
margin-bottom: 32px;
}
.summary-card {
background: #111118;
border: 1px solid #222;
border-radius: 10px;
padding: 16px;
}
.summary-card .label { font-size: 11px; color: #666; text-transform: uppercase; letter-spacing: 1px; margin-bottom: 4px; }
.summary-card .value { font-size: 28px; font-weight: 700; color: #fff; }
.summary-card .unit { font-size: 14px; color: #888; font-weight: 400; }
.summary-card.peak .value { color: #ef4444; }
.summary-card.final .value { color: #10b981; }
</style>
</head>
<body>
<h1>Waymo Perception Studio — Data Pipeline & Memory Map</h1>
<p class="subtitle">Memory usage when loading one segment (199 frames) · 9 Parquet files · 3 LiDAR Workers + 2 Camera Workers</p>
<div class="summary-cards">
<div class="summary-card">
<div class="label">Compressed on Disk</div>
<div class="value">~502 <span class="unit">MB</span></div>
</div>
<div class="summary-card peak">
<div class="label">Peak Memory (All Threads)</div>
<div class="value">~1.38 <span class="unit">GB</span></div>
</div>
<div class="summary-card final">
<div class="label">Steady-State (Cached)</div>
<div class="value">~890 <span class="unit">MB</span></div>
</div>
<div class="summary-card">
<div class="label">Frames × Sensors</div>
<div class="value">199 <span class="unit">× 5</span></div>
</div>
</div>
<div class="legend">
<div class="legend-item"><div class="legend-dot" style="background:#2d5aa0"></div> Fetch / File I/O</div>
<div class="legend-item"><div class="legend-dot" style="background:#8b5cf6"></div> BROTLI Decompress</div>
<div class="legend-item"><div class="legend-dot" style="background:#f59e0b"></div> Parse (Parquet→Objects)</div>
<div class="legend-item"><div class="legend-dot" style="background:#ef4444"></div> Spherical→Cartesian conversion</div>
<div class="legend-item"><div class="legend-dot" style="background:#10b981"></div> Cache / Transfer</div>
<div class="legend-item"><div class="legend-dot" style="background:#06b6d4"></div> Worker Init</div>
<div class="legend-item"><div class="legend-dot" style="background:#f472b6"></div> Three.js / GPU Upload</div>
</div>
<!-- ============================================================ -->
<!-- PHASE 1: Open Parquet Files -->
<!-- ============================================================ -->
<div class="phase active">
<div class="phase-header">
<span class="phase-num">Phase 1</span>
<span class="phase-title">Open Parquet Files (Footer Only)</span>
<span class="phase-time">~50ms</span>
</div>
<div class="note">
<strong>9 files:</strong> only footers are read (range requests); payloads are not loaded.
Minimal memory impact: metadata only (~50 KB total).
</div>
<div class="swim-lanes">
<div class="swim-lane">
<div class="lane-label">Main Thread</div>
<div class="lane-bar-container">
<div class="lane-bar c-fetch" style="left:0; width:85%">
9× openParquetFile() — footer parse only
<span class="mem-tag">~50 KB</span>
</div>
</div>
</div>
</div>
<br>
<table class="mem-table">
<tr><th>Component</th><th>File Size (Compressed)</th><th>Footer Read</th><th>Memory</th></tr>
<tr><td>vehicle_pose</td><td>~30 KB</td><td>~1 KB</td><td class="mem-val">~5 KB</td></tr>
<tr><td>lidar_calibration</td><td>~5 KB</td><td>~1 KB</td><td class="mem-val">~3 KB</td></tr>
<tr><td>camera_calibration</td><td>~9 KB</td><td>~1 KB</td><td class="mem-val">~4 KB</td></tr>
<tr><td>lidar_box</td><td>~1.9 MB</td><td>~2 KB</td><td class="mem-val">~8 KB</td></tr>
<tr><td>camera_box</td><td>~450 KB</td><td>~2 KB</td><td class="mem-val">~6 KB</td></tr>
<tr><td>camera_to_lidar_box_association</td><td>~26 KB</td><td>~1 KB</td><td class="mem-val">~3 KB</td></tr>
<tr><td>lidar</td><td>~170 MB</td><td>~4 KB</td><td class="mem-val">~10 KB</td></tr>
<tr><td>camera_image</td><td>~330 MB</td><td>~4 KB</td><td class="mem-val">~10 KB</td></tr>
<tr><td>stats</td><td>~24 KB</td><td>~1 KB</td><td class="mem-val">~3 KB</td></tr>
</table>
</div>
<!-- ============================================================ -->
<!-- PHASE 2: Load Startup Data (Small Files) -->
<!-- ============================================================ -->
<div class="phase active">
<div class="phase-header">
<span class="phase-num">Phase 2</span>
<span class="phase-title">loadStartupData() — Small Files Full Read</span>
<span class="phase-time">~200-400ms</span>
</div>
<div class="note">
<strong>Sequential loading:</strong> readAllRows() loads 7 small files. Each goes through BROTLI decompression → Parquet parsing → JS objects.
lidar_box is the largest: ~1.9 MB compressed → ~8 MB decompressed.
</div>
<div class="swim-lanes">
<div class="swim-lane">
<div class="lane-label">vehicle_pose</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:0; width:12%">
BROTLI
<span class="mem-tag">30KB→200KB</span>
</div>
<div class="lane-bar c-parse" style="left:13%; width:15%">
readAllRows → 199 rows + buildFrameIndex + poseByFrameIndex
<span class="mem-tag">~600 KB retained</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label">lidar_calibration</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:29%; width:4%">
<span class="mem-tag">5KB→15KB</span>
</div>
<div class="lane-bar c-parse" style="left:34%; width:4%">
5 calibrations
<span class="mem-tag">~10 KB</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label">camera_calibration</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:39%; width:4%">
<span class="mem-tag">9KB→25KB</span>
</div>
<div class="lane-bar c-parse" style="left:44%; width:4%">
5 calibrations
<span class="mem-tag">~15 KB</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label">lidar_box ★</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:49%; width:12%">
BROTLI 1.9MB
<span class="mem-tag">→ ~8 MB</span>
</div>
<div class="lane-bar c-parse" style="left:62%; width:14%">
~3000 boxes + trajectory index + groupIndexBy
<span class="mem-tag">~12 MB retained</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label">camera_box</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:77%; width:6%">
450KB
<span class="mem-tag">→ ~2 MB</span>
</div>
<div class="lane-bar c-parse" style="left:84%; width:6%">
~5000 boxes + groupIndexBy
<span class="mem-tag">~4 MB</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label">association</div>
<div class="lane-bar-container">
<div class="lane-bar c-parse" style="left:91%; width:4%">
cam↔laser maps
<span class="mem-tag">~200 KB</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label">stats</div>
<div class="lane-bar-container">
<div class="lane-bar c-parse" style="left:96%; width:3%">
1 row meta
<span class="mem-tag">~5 KB</span>
</div>
</div>
</div>
</div>
<div class="note" style="margin-top:12px">
<strong>Main-thread memory at the end of Phase 2: ~17 MB</strong><br>
(vehicle_pose 600KB + lidar_box 12MB + camera_box 4MB + assoc 200KB + calibrations 25KB + stats 5KB)
</div>
</div>
<!-- ============================================================ -->
<!-- PHASE 3: Worker Pool Init -->
<!-- ============================================================ -->
<div class="phase active">
<div class="phase-header">
<span class="phase-num">Phase 3</span>
<span class="phase-title">Worker Pool Initialization (LiDAR ∥ Camera)</span>
<span class="phase-time">~100-300ms</span>
</div>
<div class="note">
<strong>Parallel</strong>: 3 LiDAR workers and 2 camera workers initialize concurrently.
Each worker independently opens the Parquet file and reparses its footer.
Per-worker memory overhead plus duplicated JS engine state.
</div>
<div class="parallel-bracket">
<span class="parallel-label">Promise.all — Parallel execution</span>
<div class="swim-lanes" style="padding-top:8px">
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-worker">LiDAR W1</span></div>
<div class="lane-bar-container">
<div class="lane-bar c-worker" style="left:0; width:80%">
openParquetFile('lidar') + buildHeavyFileFrameIndex
<span class="mem-tag">~5 MB (Worker heap + metadata + key index)</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-worker">LiDAR W2</span></div>
<div class="lane-bar-container">
<div class="lane-bar c-worker" style="left:0; width:80%">
openParquetFile('lidar') + buildHeavyFileFrameIndex
<span class="mem-tag">~5 MB</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-worker">LiDAR W3</span></div>
<div class="lane-bar-container">
<div class="lane-bar c-worker" style="left:0; width:80%">
openParquetFile('lidar') + buildHeavyFileFrameIndex
<span class="mem-tag">~5 MB</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-cam">Camera W1</span></div>
<div class="lane-bar-container">
<div class="lane-bar c-worker" style="left:0; width:60%; background:linear-gradient(135deg,#059669,#047857)">
openParquetFile('camera_image')
<span class="mem-tag">~3 MB (Worker heap + metadata)</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-cam">Camera W2</span></div>
<div class="lane-bar-container">
<div class="lane-bar c-worker" style="left:0; width:60%; background:linear-gradient(135deg,#059669,#047857)">
openParquetFile('camera_image')
<span class="mem-tag">~3 MB</span>
</div>
</div>
</div>
</div>
</div>
<div class="note" style="margin-top:12px">
<strong>Total memory at the end of Phase 3: ~38 MB</strong><br>
Main: 17 MB + 3× LiDAR Worker: 15 MB + 2× Camera Worker: 6 MB
</div>
</div>
<!-- ============================================================ -->
<!-- PHASE 4: First 2 Row Groups (Blocking) -->
<!-- ============================================================ -->
<div class="phase active">
<div class="phase-header">
<span class="phase-num">Phase 4</span>
<span class="phase-title">First 2 Row Groups — LiDAR + Camera (Blocking)</span>
<span class="phase-time">~800-1500ms</span>
</div>
<div class="note">
<strong>⚡ Peak Memory Interval</strong>. LiDAR RG0+RG1 and camera RG0+RG1 are processed concurrently.
Each LiDAR RG contains ~51 frames × 5 sensors = ~255 range images.
BROTLI decompression: ~42 MB compressed → ~100 MB decompressed per RG (temporary).
</div>
<div class="parallel-bracket">
<span class="parallel-label">Promise.all — LiDAR RG0, RG1 ∥ Camera RG0, RG1</span>
<div class="swim-lanes" style="padding-top:8px">
<!-- LiDAR RG0 on Worker 1 -->
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-worker">LiDAR W1</span> RG0</div>
<div class="lane-bar-container">
<div class="lane-bar c-fetch" style="left:0; width:10%">
fetch
<span class="mem-tag">~42 MB</span>
</div>
<div class="lane-bar c-decomp" style="left:11%; width:25%">
BROTLI decompress (~255 rows)
<span class="mem-tag">→ ~100 MB transient</span>
</div>
<div class="lane-bar c-parse" style="left:37%; width:10%">
parse rows
<span class="mem-tag">~60 MB</span>
</div>
<div class="lane-bar c-convert" style="left:48%; width:25%">
spherical→xyz (51 frames × 5 sensors)
<span class="mem-tag">~24 MB Float32</span>
</div>
<div class="lane-bar c-cache" style="left:74%; width:10%">
transfer→main
<span class="mem-tag">→ 0 (transferred)</span>
</div>
</div>
</div>
<!-- LiDAR RG1 on Worker 2 -->
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-worker">LiDAR W2</span> RG1</div>
<div class="lane-bar-container">
<div class="lane-bar c-fetch" style="left:0; width:10%">
fetch
<span class="mem-tag">~42 MB</span>
</div>
<div class="lane-bar c-decomp" style="left:11%; width:25%">
BROTLI decompress
<span class="mem-tag">→ ~100 MB transient</span>
</div>
<div class="lane-bar c-parse" style="left:37%; width:10%">
parse
<span class="mem-tag">~60 MB</span>
</div>
<div class="lane-bar c-convert" style="left:48%; width:25%">
spherical→xyz
<span class="mem-tag">~24 MB Float32</span>
</div>
<div class="lane-bar c-cache" style="left:74%; width:10%">
transfer
<span class="mem-tag">→ 0</span>
</div>
</div>
</div>
<!-- Camera RG0 on Camera Worker 1 -->
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-cam">Camera W1</span> RG0</div>
<div class="lane-bar-container">
<div class="lane-bar c-fetch" style="left:0; width:12%; background:linear-gradient(135deg,#2d7a5a,#1e5a3f)">
fetch
<span class="mem-tag">~80 MB</span>
</div>
<div class="lane-bar c-decomp" style="left:13%; width:30%; background:linear-gradient(135deg,#059669,#047857)">
BROTLI → JPEG extract (50 frames × 5 cams)
<span class="mem-tag">~85 MB transient</span>
</div>
<div class="lane-bar c-cache" style="left:44%; width:8%">
transfer JPEGs
<span class="mem-tag">→ 0</span>
</div>
</div>
</div>
<!-- Camera RG1 on Camera Worker 2 -->
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-cam">Camera W2</span> RG1</div>
<div class="lane-bar-container">
<div class="lane-bar c-fetch" style="left:0; width:12%; background:linear-gradient(135deg,#2d7a5a,#1e5a3f)">
fetch
<span class="mem-tag">~80 MB</span>
</div>
<div class="lane-bar c-decomp" style="left:13%; width:30%; background:linear-gradient(135deg,#059669,#047857)">
BROTLI → JPEG extract
<span class="mem-tag">~85 MB transient</span>
</div>
<div class="lane-bar c-cache" style="left:44%; width:8%">
transfer
<span class="mem-tag">→ 0</span>
</div>
</div>
</div>
<!-- Main thread receiving -->
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-main">Main Thread</span></div>
<div class="lane-bar-container">
<div class="lane-bar c-cache" style="left:74%; width:24%">
cacheRowGroupFrames + cacheCameraRowGroupFrames (frameCache + cameraImageCache)
<span class="mem-tag">+48 MB lidar + 160 MB JPEG → Main</span>
</div>
</div>
</div>
</div>
</div>
<div class="note" style="margin-top:12px; border-left-color:#ef4444">
<strong style="color:#ef4444">🔴 Peak Memory Occurs When:</strong> LiDAR W1 + W2 and camera W1 + W2 decompress BROTLI concurrently<br>
<strong>Transient worker memory:</strong> 2× LiDAR (~200 MB) + 2× Camera (~170 MB) = <strong>~370 MB transient</strong><br>
<strong>Existing main-thread memory:</strong> 38 MB + incoming cached data ~100 MB<br>
<strong>→ Peak Total: ~1.38 GB</strong> (17 MB startup + 320 MB LiDAR workers + 330 MB Camera workers + 210 MB partial cache + 25 MB worker base)
</div>
</div>
<!-- ============================================================ -->
<!-- PHASE 5: First Frame Render -->
<!-- ============================================================ -->
<div class="phase active">
<div class="phase-header">
<span class="phase-num">Phase 5</span>
<span class="phase-title">First Frame Render → GPU Upload</span>
<span class="phase-time">~16ms (1 frame)</span>
</div>
<div class="swim-lanes">
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-main">Main Thread</span></div>
<div class="lane-bar-container">
<div class="lane-bar c-gpu" style="left:0; width:40%">
frameCache.get(0) → set currentFrame
<span class="mem-tag">~240 KB (1 frame points ref)</span>
</div>
<div class="lane-bar c-gpu" style="left:42%; width:55%">
Three.js BufferGeometry upload → GPU (positions + colors)
<span class="mem-tag">~2 MB VRAM (168K pts × 12 bytes)</span>
</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label">Camera Images</div>
<div class="lane-bar-container">
<div class="lane-bar c-gpu" style="left:0; width:80%">
5× JPEG → <img> src=blob: → GPU texture
<span class="mem-tag">5× ~200KB JPEG ≈ 1 MB CPU + ~15 MB decoded VRAM</span>
</div>
</div>
</div>
</div>
<div class="note">
<strong>status = 'ready'</strong>, autoplay starts and the first frame is displayed.
Adds ~17 MB GPU VRAM (point cloud + 5 camera textures).
</div>
</div>
<!-- ============================================================ -->
<!-- PHASE 6: Background Prefetch -->
<!-- ============================================================ -->
<div class="phase active">
<div class="phase-header">
<span class="phase-num">Phase 6</span>
<span class="phase-title">Background Prefetch (RG2, RG3 — remaining ~100 frames)</span>
<span class="phase-time">~2-4s (background, non-blocking)</span>
</div>
<div class="note">
LiDAR: dispatch RG2 and RG3 to WorkerPool(3); RG0 and RG1 are already complete.
Camera: dispatch RG2 and RG3 to CameraPool(2).
Workers process queued tasks as they become idle. <strong>The main thread stays free, maintaining 60 fps rendering.</strong>
</div>
<div class="parallel-bracket">
<span class="parallel-label">Background — alongside rendering</span>
<div class="swim-lanes" style="padding-top:8px">
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-worker">LiDAR W1</span> RG2</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:0; width:35%">
BROTLI + parse + convert (~51 frames)
<span class="mem-tag">~100 MB transient → 24 MB cached</span>
</div>
<div class="lane-bar c-cache" style="left:36%; width:5%">xfer</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-worker">LiDAR W2</span> RG3</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:0; width:35%">
BROTLI + parse + convert (~46 frames)
<span class="mem-tag">~100 MB transient → 22 MB cached</span>
</div>
<div class="lane-bar c-cache" style="left:36%; width:5%">xfer</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-cam">Camera W1</span> RG2</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:0; width:40%; background:linear-gradient(135deg,#059669,#047857)">
BROTLI → JPEG (~50 frames × 5 cams)
<span class="mem-tag">~85 MB transient → 80 MB JPEG cached</span>
</div>
<div class="lane-bar c-cache" style="left:41%; width:5%">xfer</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-cam">Camera W2</span> RG3</div>
<div class="lane-bar-container">
<div class="lane-bar c-decomp" style="left:0; width:40%; background:linear-gradient(135deg,#059669,#047857)">
BROTLI → JPEG (~49 frames × 5 cams)
<span class="mem-tag">~85 MB transient → 78 MB JPEG cached</span>
</div>
<div class="lane-bar c-cache" style="left:41%; width:5%">xfer</div>
</div>
</div>
<div class="swim-lane">
<div class="lane-label"><span class="thread-tag thread-main">Main Thread</span></div>
<div class="lane-bar-container">
<div class="lane-bar c-gpu" style="left:0; width:95%; background: linear-gradient(90deg, #f472b6 0%, #f472b6 50%, #10b981 50%, #10b981 100%)">
🎬 Rendering at 60 fps while received RG data populates frameCache/cameraImageCache
</div>
</div>
</div>
</div>
</div>
</div>
<!-- ============================================================ -->
<!-- Memory Timeline Chart (SVG) -->
<!-- ============================================================ -->
<div class="memory-chart">
<div class="chart-title">Memory Timeline — All threads combined (Main + 3 LiDAR Workers + 2 Camera Workers)</div>
<svg class="chart-svg" viewBox="0 0 1000 380" preserveAspectRatio="xMidYMid meet">
<!-- Grid -->
<defs>
<linearGradient id="areaGrad" x1="0" y1="0" x2="0" y2="1">
<stop offset="0%" stop-color="#ef4444" stop-opacity="0.3"/>
<stop offset="100%" stop-color="#ef4444" stop-opacity="0.02"/>
</linearGradient>
<linearGradient id="steadyGrad" x1="0" y1="0" x2="0" y2="1">
<stop offset="0%" stop-color="#10b981" stop-opacity="0.2"/>
<stop offset="100%" stop-color="#10b981" stop-opacity="0.02"/>
</linearGradient>
</defs>
<!-- Y axis labels -->
<text x="45" y="25" fill="#666" font-size="10" text-anchor="end">1.6 GB</text>
<text x="45" y="75" fill="#666" font-size="10" text-anchor="end">1.2 GB</text>
<text x="45" y="135" fill="#666" font-size="10" text-anchor="end">800 MB</text>
<text x="45" y="195" fill="#666" font-size="10" text-anchor="end">400 MB</text>
<text x="45" y="255" fill="#666" font-size="10" text-anchor="end">200 MB</text>
<text x="45" y="315" fill="#666" font-size="10" text-anchor="end">0 MB</text>
<!-- Grid lines -->
<line x1="55" y1="20" x2="980" y2="20" stroke="#222" stroke-dasharray="3"/>
<line x1="55" y1="70" x2="980" y2="70" stroke="#222" stroke-dasharray="3"/>
<line x1="55" y1="130" x2="980" y2="130" stroke="#222" stroke-dasharray="3"/>
<line x1="55" y1="190" x2="980" y2="190" stroke="#222" stroke-dasharray="3"/>
<line x1="55" y1="250" x2="980" y2="250" stroke="#222" stroke-dasharray="3"/>
<line x1="55" y1="310" x2="980" y2="310" stroke="#444"/>
<!-- X axis time labels -->
<text x="55" y="340" fill="#666" font-size="10">0s</text>
<text x="155" y="340" fill="#666" font-size="10">Phase 1</text>
<text x="255" y="340" fill="#666" font-size="10">Phase 2</text>
<text x="405" y="340" fill="#666" font-size="10">Phase 3</text>
<text x="530" y="340" fill="#666" font-size="10">Phase 4</text>
<text x="720" y="340" fill="#666" font-size="10">1st Frame</text>
<text x="830" y="340" fill="#666" font-size="10">Prefetch</text>
<text x="950" y="340" fill="#666" font-size="10">Steady</text>
<!-- Phase divider lines -->
<line x1="140" y1="15" x2="140" y2="315" stroke="#333" stroke-dasharray="2"/>
<line x1="240" y1="15" x2="240" y2="315" stroke="#333" stroke-dasharray="2"/>
<line x1="390" y1="15" x2="390" y2="315" stroke="#333" stroke-dasharray="2"/>
<line x1="500" y1="15" x2="500" y2="315" stroke="#333" stroke-dasharray="2"/>
<line x1="710" y1="15" x2="710" y2="315" stroke="#333" stroke-dasharray="2"/>
<line x1="810" y1="15" x2="810" y2="315" stroke="#333" stroke-dasharray="2"/>
<!-- Memory area (total across all threads) -->
<!-- Path: Phase1(~0.05) → Phase2(~17MB) → Phase3(~38) → Phase4 peak(~1570MB) → drop to ~720 steady -->
<path d="
M 55,310
L 140,309
L 240,307
L 390,303
L 500,300
L 560,200
L 610,80
L 650,22
L 680,50
L 710,100
L 750,130
L 810,135
L 870,155
L 920,170
L 960,175
L 980,177
L 980,310
Z
" fill="url(#areaGrad)"/>
<!-- Total memory line -->
<path d="
M 55,310
L 140,309
L 240,307
L 390,303
L 500,300
L 560,200
L 610,80
L 650,22
L 680,50
L 710,100
L 750,130
L 810,135
L 870,155
L 920,170
L 960,175
L 980,177
" fill="none" stroke="#ef4444" stroke-width="2.5"/>
<!-- Steady state (retained) line -->
<path d="
M 55,310
L 140,310
L 240,307
L 390,303
L 500,300
L 560,260
L 610,230
L 650,200
L 680,185
L 710,177
L 750,177
L 810,177
L 870,177
L 920,177
L 960,177
L 980,177
" fill="none" stroke="#10b981" stroke-width="1.5" stroke-dasharray="5,3"/>
<!-- Peak annotation -->
<circle cx="650" cy="22" r="5" fill="#ef4444"/>
<line x1="650" y1="22" x2="690" y2="46" stroke="#ef4444" stroke-width="1"/>
<rect x="690" y="36" width="110" height="22" rx="4" fill="#1a1a24" stroke="#ef4444" stroke-width="1"/>
<text x="745" y="51" fill="#ef4444" font-size="11" font-weight="700" text-anchor="middle">Peak ~1.38 GB</text>
<!-- Steady annotation -->
<circle cx="960" cy="177" r="4" fill="#10b981"/>
<rect x="870" y="160" width="85" height="18" rx="3" fill="#1a1a24" stroke="#10b981" stroke-width="1"/>
<text x="912" y="173" fill="#10b981" font-size="10" font-weight="700" text-anchor="middle">Steady ~890MB</text>
<!-- Worker transient annotations -->
<rect x="520" y="60" width="90" height="16" rx="3" fill="#8b5cf630" stroke="#8b5cf6" stroke-width="1"/>
<text x="565" y="72" fill="#a78bfa" font-size="9" font-weight="600" text-anchor="middle">Workers transient</text>
<!-- Legend in chart -->
<circle cx="60" cy="360" r="4" fill="#ef4444"/>
<text x="68" y="364" fill="#999" font-size="10">Total (peak)</text>
<line x1="140" y1="360" x2="160" y2="360" stroke="#10b981" stroke-width="1.5" stroke-dasharray="5,3"/>
<text x="165" y="364" fill="#999" font-size="10">Retained (steady state)</text>
</svg>
</div>
<!-- ============================================================ -->
<!-- Detailed Memory Breakdown Table -->
<!-- ============================================================ -->
<div class="memory-chart" style="margin-top:24px">
<div class="chart-title">Steady-State Memory Breakdown (after all 199 frames are cached)</div>
<table class="mem-table">
<tr>
<th>Data Store</th>
<th>Thread</th>
<th>Items</th>
<th>Per-Item Size</th>
<th>Total Retained</th>
<th>Notes</th>
</tr>
<tr>
<td><strong>frameCache</strong><br>(LiDAR point clouds)</td>
<td><span class="thread-tag thread-main">Main</span></td>
<td>199 frames × 5 sensors</td>
<td>~168K pts × 24B (POINT_STRIDE=6 × 4B) = <strong>~4 MB/frame</strong><br>
<em style="color:#888;font-size:11px">Note: subarray().buffer transfers the full allocation, larger than valid data</em></td>
<td class="mem-val">~500 MB</td>
<td>Float32Array transferred from workers (zero-copy via Transferable). Actual allocation = maxPoints × STRIDE (including empty pixels)</td>
</tr>
<tr>
<td><strong>cameraImageCache</strong><br>(JPEG blobs)</td>
<td><span class="thread-tag thread-main">Main</span></td>
<td>199 frames × 5 cameras</td>
<td>~300 KB/image (JPEG compressed)</td>
<td class="mem-val">~300 MB</td>
<td>Raw JPEG ArrayBuffers — 330 MB Parquet minus ~10% Parquet overhead</td>
</tr>
<tr>
<td><strong>lidarBoxByFrame</strong><br>(3D bounding boxes)</td>
<td><span class="thread-tag thread-main">Main</span></td>
<td>~3000 boxes (199 frames)</td>
<td>~4 KB/box (14 fields)</td>
<td class="mem-val">~12 MB</td>
<td>JS objects with center, size, heading, type, id</td>
</tr>
<tr>
<td><strong>cameraBoxByFrame</strong><br>(2D bounding boxes)</td>
<td><span class="thread-tag thread-main">Main</span></td>
<td>~5000 boxes (199 frames)</td>
<td>~800 B/box</td>
<td class="mem-val">~4 MB</td>
<td>center_x, center_y, width, height, type per cam</td>
</tr>
<tr>
<td><strong>objectTrajectories</strong></td>
<td><span class="thread-tag thread-main">Main</span></td>
<td>~100 unique objects</td>
<td>~199 entries × 40B</td>
<td class="mem-val">~0.8 MB</td>
<td>frameIndex, x, y, z, type per tracking point</td>
</tr>
<tr>
<td><strong>vehiclePoseByFrame</strong><br>+ poseByFrameIndex</td>
<td><span class="thread-tag thread-main">Main</span></td>
<td>199 frames</td>
<td>16 floats × 8B = 128B</td>
<td class="mem-val">~0.6 MB</td>
<td>4×4 row-major transform matrices</td>
</tr>
<tr>
<td><strong>association maps</strong></td>
<td><span class="thread-tag thread-main">Main</span></td>
<td>~3000 entries</td>
<td>~100B/entry</td>
<td class="mem-val">~0.3 MB</td>
<td>cam↔laser ID string maps</td>
</tr>
<tr>
<td><strong>Worker heap overhead</strong><br>(3 LiDAR + 2 Camera)</td>
<td><span class="thread-tag thread-worker">Workers</span></td>
<td>5 workers idle</td>
<td>~5-15 MB/worker (JS engine + modules)</td>
<td class="mem-val">~50 MB</td>
<td>Workers stay alive for segment switching</td>
</tr>
<tr>
<td><strong>Three.js / GPU refs</strong></td>
<td><span class="thread-tag thread-main">Main</span></td>
<td>1 active frame</td>
<td>BufferGeometry + textures</td>
<td class="mem-val">~20 MB</td>
<td>CPU-side refs; GPU VRAM separate (~17 MB)</td>
</tr>
<tr style="background:#14141e; font-weight:700">
<td colspan="4" style="text-align:right; color:#10b981">TOTAL STEADY-STATE →</td>
<td class="mem-val mem-retained" style="font-size:16px">~890 MB</td>
<td></td>
</tr>
</table>
</div>
<!-- ============================================================ -->
<!-- Peak Memory Detail -->
<!-- ============================================================ -->
<div class="memory-chart" style="margin-top:24px">
<div class="chart-title">Peak Memory Detail (Phase 4 — 2× LiDAR + 2× Camera RG processed concurrently)</div>
<table class="mem-table">
<tr>
<th>Location</th>
<th>What</th>
<th>Memory</th>
<th>Lifetime</th>
</tr>
<tr>
<td><span class="thread-tag thread-main">Main</span></td>
<td>Phase 2 retained (boxes, poses, calibrations, metadata)</td>
<td class="mem-val">~17 MB</td>
<td>Permanent</td>
</tr>
<tr>
<td><span class="thread-tag thread-worker">LiDAR W1</span></td>
<td>RG0: fetch buffer (compressed) + BROTLI output + range image arrays</td>
<td class="mem-val mem-peak">~160 MB</td>
<td>Transient (GC after transfer)</td>
</tr>
<tr>
<td><span class="thread-tag thread-worker">LiDAR W2</span></td>
<td>RG1: fetch buffer + BROTLI output + range image arrays</td>
<td class="mem-val mem-peak">~160 MB</td>
<td>Transient (GC after transfer)</td>
</tr>
<tr>
<td><span class="thread-tag thread-worker">LiDAR W3</span></td>
<td>Idle (base heap only)</td>
<td class="mem-val">~5 MB</td>
<td>Permanent</td>
</tr>
<tr>
<td><span class="thread-tag thread-cam">Camera W1</span></td>
<td>RG0: compressed fetch + BROTLI output (~50 frames × 5 imgs)</td>
<td class="mem-val mem-peak">~165 MB</td>
<td>Transient (GC after transfer)</td>
</tr>
<tr>
<td><span class="thread-tag thread-cam">Camera W2</span></td>
<td>RG1: compressed fetch + BROTLI output</td>
<td class="mem-val mem-peak">~165 MB</td>
<td>Transient (GC after transfer)</td>
</tr>
<tr>
<td><span class="thread-tag thread-main">Main</span></td>
<td>frameCache (partially filled — RG0 data arriving)</td>
<td class="mem-val">~130 MB</td>
<td>Permanent (growing)</td>
</tr>