Gahow Wang
9c105cf05a
Reuse and concurrency axes redone with proper controlled variables, plus
the orchestration used to run them on dash0:
- run_reuse_fixed.sh: hold REAL prefill work (delta) constant, vary only
cached prefix -> reuse = C/(C+U). Supersedes old fig1 (which held
input=8192 and sliced prefix out, confounding "more reuse" with "less
prefill").
- run_conc.sh: agentic-corner config (in=32768, delta=512, reuse=0.984,
out=128) that exposes PD's structural KV-transfer tax. Supersedes old fig3.
- run_campaign{,2,3}.sh, backfill_d2048o128.sh: serial campaign drivers
(strictly one driver at a time), out=128 sweeps, PD wall-cap for
collapse-draining high-reuse arms, and flaked-arm backfill.
- mb5_run_gpu.sh: per-config bring-up / replay / teardown orchestrator.
- plot_pd_crossover.py: render the reuse_compare figures from fig_agg dumps.
- fig_agg.py: tolerate null stats from fully-collapsed arms (0 successes
write the stat keys as null; `dict.get(k, {})` returns null, not {}).
Data: fig1_reuse_fixed.json, fig1_reuse_d{1024,2048}_o128.json
Figs: reuse_compare_AB.png, reuse_compare_ABC.png
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2 lines
9.3 KiB
JSON
2 lines
9.3 KiB
JSON
[{"name": "reuse_p18432_d1024_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 2.011140084010549, "e2e_p90": 6.595956567587561, "e2e_p99": 13.843295609399851, "e2e_mean": 3.1182714335319965, "ttft_p90": 5.735121128271566, "tpot_p99": 0.007805189969399431, "tps": 274.4217808266656, "wall": 179.11114727094537, "pu": 41.73809523809524, "du": 24.771825396825395, "apc": 0.8289473684210527}, {"name": "reuse_p18432_d1024_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 1.2469783920387272, "e2e_p90": 3.5110186243895436, "e2e_p99": 5.371424139904558, "e2e_mean": 1.8077833487527641, "ttft_p90": 2.6449656961078287, "tpot_p99": 0.008740999029887894, "tps": 427.0965500477582, "wall": 115.0840483129723, "pu": 31.495454545454546, "du": 58.32272727272727, "apc": 0.8289473684210527}, {"name": "reuse_p18432_d1024_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 1.5855289729661308, "e2e_p90": 3.5929854729911312, "e2e_p99": 5.639228541038465, "e2e_mean": 1.9572647203356912, "ttft_p90": 2.3784743072232226, "tpot_p99": 0.014603613033605647, "tps": 396.0826187284008, "wall": 124.09532172302715, "pu": 21.093220338983052, "du": 82.70338983050847, "apc": 0.8289473684210527}, {"name": "reuse_p18432_d1024_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 1.3775086904934142, "e2e_p90": 3.1152419353893483, "e2e_p99": 7.911835642378456, "e2e_mean": 1.8861664114436583, "ttft_p90": 1.989750977011863, "tpot_p99": 0.04218796803300668, "tps": 419.48032182196613, "wall": 117.17355366400443, "pu": 41.058035714285715, "du": null, "apc": 0.8289473684210527}, {"name": "reuse_p2048_d1024_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 0.792972255992936, "e2e_p90": 0.9489053479628637, "e2e_p99": 1.637123176814641, "e2e_mean": 0.8399674154458504, "ttft_p90": 0.25993717868113886, "tpot_p99": 0.005669725780525634, "tps": 791.8028909214946, "wall": 62.07605524500832, "pu": 31.716666666666665, "du": 66.96666666666667, "apc": 0.5833333333333334}, {"name": "reuse_p2048_d1024_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 0.8240640689909924, "e2e_p90": 0.9465674142818898, "e2e_p99": 1.4786975827836435, "e2e_mean": 0.8572143532189026, "ttft_p90": 0.23996333410032095, "tpot_p99": 0.006090612034991707, "tps": 783.216723293919, "wall": 62.75657622999279, "pu": 15.566666666666666, "du": 82.11666666666666, "apc": 0.5833333333333334}, {"name": "reuse_p2048_d1024_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 0.9896964269864839, "e2e_p90": 1.1497688122093677, "e2e_p99": 1.787693639339414, "e2e_mean": 1.0255024050854142, "ttft_p90": 0.2459226590231992, "tpot_p99": 0.008264087546006299, "tps": 691.4021770340051, "wall": 71.09031708701514, "pu": 8.740196078431373, "du": 95.52941176470588, "apc": 0.5833333333333334}, {"name": "reuse_p2048_d1024_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 0.9049637360149063, "e2e_p90": 1.2046391166280954, "e2e_p99": 1.5656454849703014, "e2e_mean": 0.9364410856549815, "ttft_p90": 0.2209109389339573, "tpot_p99": 0.010840669604821322, "tps": 697.8220171948692, "wall": 70.43629863898968, "pu": 39.75735294117647, "du": null, "apc": 0.5833333333333334}, {"name": "reuse_p38912_d1024_o128_2P+6D_rep1", "arm": "2P+6D", "n": 348, "req": 384, "e2e_p50": 5.966376932017738, "e2e_p90": 41.55034995879979, "e2e_p99": 67.95094630374655, "e2e_mean": 17.63732494259318, "ttft_p90": 40.65220642983333, "tpot_p99": 0.00965455187338873, "tps": 31.074994331197068, "wall": 1433.4354988210252, "pu": 29.08110119047619, "du": 3.8563988095238093, "apc": 0.003541386023109088}, {"name": "reuse_p38912_d1024_o128_4P+4D_rep1", "arm": "4P+4D", "n": 331, "req": 384, "e2e_p50": 1.8283151049981825, "e2e_p90": 21.967222570965532, "e2e_p99": 463.6285565488564, "e2e_mean": 32.049094769653685, "ttft_p90": 20.93402343802154, "tpot_p99": 0.010148854255877936, "tps": 22.304023016957444, "wall": 1899.5676236429717, "pu": 6.330527497194164, "du": 4.0412457912457915, "apc": 0.3941685392519007}, {"name": "reuse_p38912_d1024_o128_6P+2D_rep1", "arm": "6P+2D", "n": 331, "req": 384, "e2e_p50": 1.5403626519837417, "e2e_p90": 8.17470415995922, "e2e_p99": 15.978965973359298, "e2e_mean": 3.262189171190994, "ttft_p90": 7.065585554984864, "tpot_p99": 0.018555945050481002, "tps": 46.7175532998922, "wall": 906.8968087440007, "pu": 7.588627450980392, "du": 14.944705882352942, "apc": 0.8538197197991012}, {"name": "reuse_p38912_d1024_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 2.0001139829983003, "e2e_p90": 7.92033636061824, "e2e_p99": 17.198196112165586, "e2e_mean": 3.6551371743582117, "ttft_p90": 6.314943998103264, "tpot_p99": 0.11950804649024022, "tps": 240.1072077908529, "wall": 204.70855686604045, "pu": 41.98711340206186, "du": null, "apc": 0.8525641025641025}, {"name": "reuse_p4096_d1024_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 0.8382689229911193, "e2e_p90": 1.1122795563715044, "e2e_p99": 2.1025787757901724, "e2e_mean": 0.9100473551948198, "ttft_p90": 0.44005942158401007, "tpot_p99": 0.005548434369893942, "tps": 741.2065308688034, "wall": 66.31350096495589, "pu": 29.515625, "du": 64.26041666666667, "apc": 0.7}, {"name": "reuse_p4096_d1024_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 0.845705887011718, "e2e_p90": 1.0994400390714874, "e2e_p99": 1.7404719427420072, "e2e_mean": 0.9025246042629078, "ttft_p90": 0.3965871149906888, "tpot_p99": 0.00621245443521207, "tps": 753.8055854507271, "wall": 65.20514168200316, "pu": 17.241935483870968, "du": 89.73387096774194, "apc": 0.7}, {"name": "reuse_p4096_d1024_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 1.041008985514054, "e2e_p90": 1.2969930486229715, "e2e_p99": 2.009561244166107, "e2e_mean": 1.0891431775930869, "ttft_p90": 0.4054622187104541, "tpot_p99": 0.007866649039696508, "tps": 662.7457211326365, "wall": 74.16419062804198, "pu": 10.566666666666666, "du": 97.14285714285714, "apc": 0.7}, {"name": "reuse_p4096_d1024_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 0.9978507100022398, "e2e_p90": 1.376642303710105, "e2e_p99": 1.8764825401920837, "e2e_mean": 1.0177971177561933, "ttft_p90": 0.3318765960051678, "tpot_p99": 0.012630984430049199, "tps": 663.862182067196, "wall": 74.03946380398702, "pu": 38.21071428571429, "du": null, "apc": 0.7}, {"name": "reuse_p512_d1024_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 0.7468664619955234, "e2e_p90": 0.8154127431334928, "e2e_p99": 1.4249815517506803, "e2e_mean": 0.7755074385066413, "ttft_p90": 0.15653720167465504, "tpot_p99": 0.005882820960814427, "tps": 836.3744558932802, "wall": 58.76793540699873, "pu": 22.75, "du": 68.125, "apc": 0.2916666666666667}, {"name": "reuse_p512_d1024_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 0.8027036284911446, "e2e_p90": 0.8790760690986644, "e2e_p99": 1.2941429702786267, "e2e_mean": 0.8199292982858424, "ttft_p90": 0.14772663660696708, "tpot_p99": 0.006057956323456606, "tps": 805.372620261493, "wall": 61.03013532300247, "pu": 13.181034482758621, "du": 76.17241379310344, "apc": 0.2916666666666667}, {"name": "reuse_p512_d1024_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 0.9647596840222832, "e2e_p90": 1.053520446305629, "e2e_p99": 1.528927146314527, "e2e_mean": 0.9760259314924346, "ttft_p90": 0.15018326903227716, "tpot_p99": 0.007652870517977565, "tps": 717.9133965889113, "wall": 68.46508260403061, "pu": 6.55050505050505, "du": 94.60606060606061, "apc": 0.2916666666666667}, {"name": "reuse_p512_d1024_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 0.8443412214983255, "e2e_p90": 1.1002090520923957, "e2e_p99": 1.4714346799789941, "e2e_mean": 0.8751813448542028, "ttft_p90": 0.18448241882724686, "tpot_p99": 0.009217632268207137, "tps": 737.9202829911662, "wall": 66.60882094304543, "pu": 39.22265625, "du": null, "apc": 0.2916666666666667}, {"name": "reuse_p8192_d1024_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 0.9952880369964987, "e2e_p90": 1.638875626219669, "e2e_p99": 3.3932996814849328, "e2e_mean": 1.182382257353917, "ttft_p90": 0.9387523760029581, "tpot_p99": 0.00642865186423342, "tps": 614.0489008999317, "wall": 80.04574216803303, "pu": 48.36842105263158, "du": 49.6578947368421, "apc": 0.7777777777777778}, {"name": "reuse_p8192_d1024_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 0.9673103050154168, "e2e_p90": 1.6615959092101547, "e2e_p99": 2.7100385388941506, "e2e_mean": 1.122088482831714, "ttft_p90": 0.8661944593768566, "tpot_p99": 0.007023902694157537, "tps": 636.3849652635973, "wall": 77.23626842699014, "pu": 26.216216216216218, "du": 73.3108108108108, "apc": 0.7777777777777778}, {"name": "reuse_p8192_d1024_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 1.1928778715082444, "e2e_p90": 1.806718060915591, "e2e_p99": 2.9189987593132543, "e2e_mean": 1.307137680928766, "ttft_p90": 0.8438458036922384, "tpot_p99": 0.009671045643947548, "tps": 571.1298010568557, "wall": 86.06099683302455, "pu": 13.861788617886178, "du": 94.15853658536585, "apc": 0.7777777777777778}, {"name": "reuse_p8192_d1024_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 1.1058233095391188, "e2e_p90": 2.0377895820070986, "e2e_p99": 2.6251585837343008, "e2e_mean": 1.2416493104550075, "ttft_p90": 0.7047733141400385, "tpot_p99": 0.018370147039870898, "tps": 565.7217537746836, "wall": 86.88370152295101, "pu": 41.207317073170735, "du": null, "apc": 0.7777777777777778}]
|