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
8.8 KiB
JSON
2 lines
8.8 KiB
JSON
[{"name": "reuse_p18432_d2048_o128_2P+6D_rep1", "arm": "2P+6D", "n": 0, "req": 384, "e2e_p50": null, "e2e_p90": null, "e2e_p99": null, "e2e_mean": null, "ttft_p90": null, "tpot_p99": null, "tps": 0.0, "wall": 500.0238099139533, "pu": 7.442796610169491, "du": 0.0, "apc": 0.5504807692307693}, {"name": "reuse_p18432_d2048_o128_4P+4D_rep1", "arm": "4P+4D", "n": 0, "req": 384, "e2e_p50": null, "e2e_p90": null, "e2e_p99": null, "e2e_mean": null, "ttft_p90": null, "tpot_p99": null, "tps": 0.0, "wall": 500.0241022160044, "pu": 7.969279661016949, "du": 0.0, "apc": 0.4124119304837952}, {"name": "reuse_p18432_d2048_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 1.597727442975156, "e2e_p90": 3.8423233716050165, "e2e_p99": 5.946185397276423, "e2e_mean": 2.1024188470485874, "ttft_p90": 2.594332438148558, "tpot_p99": 0.014465094759590983, "tps": 378.11492013389045, "wall": 129.9922255979618, "pu": 26.701612903225808, "du": 81.08064516129032, "apc": 0.7875}, {"name": "reuse_p18432_d2048_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 1.7018155964906327, "e2e_p90": 3.876369076193077, "e2e_p99": 7.142674320546252, "e2e_mean": 2.1840314005736823, "ttft_p90": 2.1786344624473712, "tpot_p99": 0.03983575570474082, "tps": 369.2181586612368, "wall": 133.12454668595456, "pu": 42.442460317460316, "du": null, "apc": 0.7875}, {"name": "reuse_p2048_d2048_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 0.8702527615241706, "e2e_p90": 1.0261618579854257, "e2e_p99": 1.7815040894789855, "e2e_mean": 0.9216887839597803, "ttft_p90": 0.37184600940672674, "tpot_p99": 0.00531729997127449, "tps": 739.561895750781, "wall": 66.46096869296161, "pu": 42.046875, "du": 64.90104166666667, "apc": 0.4375}, {"name": "reuse_p2048_d2048_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 0.9034653444832657, "e2e_p90": 1.0732842879777307, "e2e_p99": 1.7153989363042639, "e2e_mean": 0.9481709641428703, "ttft_p90": 0.3279811007028911, "tpot_p99": 0.006350442920380646, "tps": 727.0453582087905, "wall": 67.6051355600357, "pu": 24.71875, "du": 78.4140625, "apc": 0.4375}, {"name": "reuse_p2048_d2048_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 1.0671798489929643, "e2e_p90": 1.2583458551147486, "e2e_p99": 1.8928272985597145, "e2e_mean": 1.1038180056437643, "ttft_p90": 0.3353222259087488, "tpot_p99": 0.008093095353775196, "tps": 658.7111050415339, "wall": 74.61844748602016, "pu": 15.444444444444445, "du": 96.48611111111111, "apc": 0.4375}, {"name": "reuse_p2048_d2048_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 1.0099046650284436, "e2e_p90": 1.403463970933808, "e2e_p99": 1.9772926643601396, "e2e_mean": 1.0592681981953926, "ttft_p90": 0.3512645903218072, "tpot_p99": 0.011878674168755692, "tps": 637.5511369696022, "wall": 77.09499230701476, "pu": 38.972972972972975, "du": null, "apc": 0.4375}, {"name": "reuse_p38912_d2048_o128_2P+6D_rep1", "arm": "2P+6D", "n": 0, "req": 384, "e2e_p50": null, "e2e_p90": null, "e2e_p99": null, "e2e_mean": null, "ttft_p90": null, "tpot_p99": null, "tps": 0.0, "wall": 500.109733818972, "pu": 8.895744680851063, "du": 0.0, "apc": 0.7069767441860465}, {"name": "reuse_p38912_d2048_o128_4P+4D_rep1", "arm": "4P+4D", "n": 19, "req": 384, "e2e_p50": 8.589444032986648, "e2e_p90": 27.915465352416504, "e2e_p99": 28.366348578078906, "e2e_mean": 13.736264503263795, "ttft_p90": 26.597048029780854, "tpot_p99": 0.012102746355822164, "tps": 4.86343652912101, "wall": 500.0579292929615, "pu": 4.287234042553192, "du": 1.3297872340425532, "apc": 0.76}, {"name": "reuse_p38912_d2048_o128_6P+2D_rep1", "arm": "6P+2D", "n": 367, "req": 384, "e2e_p50": 2.2283155800541863, "e2e_p90": 9.522073879209366, "e2e_p99": 21.179028030608304, "e2e_mean": 4.37911018002772, "ttft_p90": 8.025956917623997, "tpot_p99": 0.018388294600606633, "tps": 97.75392348851425, "wall": 480.55360156996176, "pu": 18.72197640117994, "du": 31.74778761061947, "apc": 0.83125}, {"name": "reuse_p38912_d2048_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 2.5377507034572773, "e2e_p90": 8.48644620324485, "e2e_p99": 23.11029849225888, "e2e_mean": 4.287540712224351, "ttft_p90": 6.662501206889283, "tpot_p99": 0.12405163028898951, "tps": 202.86541371579986, "wall": 242.28871299303137, "pu": 37.78399122807018, "du": null, "apc": 0.83125}, {"name": "reuse_p4096_d2048_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 0.9403678920061793, "e2e_p90": 1.3765556774975263, "e2e_p99": 2.223834651577637, "e2e_mean": 1.0481886899786634, "ttft_p90": 0.6951776356203481, "tpot_p99": 0.005652166551770092, "tps": 676.2238276411496, "wall": 72.68599240499316, "pu": 44.628571428571426, "du": 56.67619047619048, "apc": 0.5833333333333334}, {"name": "reuse_p4096_d2048_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 0.933273567003198, "e2e_p90": 1.19254511789768, "e2e_p99": 1.9197397668258063, "e2e_mean": 0.9934309769355423, "ttft_p90": 0.48482207448687403, "tpot_p99": 0.006479399048977194, "tps": 704.4073580750398, "wall": 69.77780603303108, "pu": 28.62878787878788, "du": 86.50757575757575, "apc": 0.5833333333333334}, {"name": "reuse_p4096_d2048_o128_6P+2D_rep1", "arm": "6P+2D", "n": 82, "req": 384, "e2e_p50": 1.9051678330288269, "e2e_p90": 5.428465517004953, "e2e_p99": 6.899000353141454, "e2e_mean": 2.8002093829528327, "ttft_p90": 4.15596770622069, "tpot_p99": 0.013283818648503837, "tps": 20.99105592369273, "wall": 500.02248758496717, "pu": 1.8764124293785311, "du": 3.419491525423729, "apc": 0.5788113695090439}, {"name": "reuse_p4096_d2048_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 1.145786852488527, "e2e_p90": 1.7778903291677124, "e2e_p99": 2.219522972920677, "e2e_mean": 1.1847675088301912, "ttft_p90": 0.5011492442048621, "tpot_p99": 0.014808055440112447, "tps": 584.2097378547378, "wall": 84.13416760304244, "pu": 36.628125, "du": null, "apc": 0.5833333333333334}, {"name": "reuse_p512_d2048_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 0.8357307700207457, "e2e_p90": 0.9345466257887891, "e2e_p99": 1.6236903413617998, "e2e_mean": 0.867049211952235, "ttft_p90": 0.28156024306081234, "tpot_p99": 0.005549601508573975, "tps": 772.9495392094975, "wall": 63.590179574035574, "pu": 34.4, "du": 64.77777777777777, "apc": 0.175}, {"name": "reuse_p512_d2048_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 0.8685456210223492, "e2e_p90": 0.9515164563606959, "e2e_p99": 1.4229793156281814, "e2e_mean": 0.8936527385265739, "ttft_p90": 0.21474569473066368, "tpot_p99": 0.006214100767071836, "tps": 761.8182639550365, "wall": 64.51932478597155, "pu": 18.016129032258064, "du": 81.63709677419355, "apc": 0.175}, {"name": "reuse_p512_d2048_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 1.064892176509602, "e2e_p90": 1.153052446676884, "e2e_p99": 1.531589741804638, "e2e_mean": 1.070551099629938, "ttft_p90": 0.23582011537509967, "tpot_p99": 0.007750524568135896, "tps": 667.8236536468205, "wall": 73.60026817198377, "pu": 8.876190476190477, "du": 92.81428571428572, "apc": 0.175}, {"name": "reuse_p512_d2048_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 0.98292404046515, "e2e_p90": 1.314295787154697, "e2e_p99": 1.7231975731498097, "e2e_mean": 0.9991093327666931, "ttft_p90": 0.34203003942966437, "tpot_p99": 0.010019689613378492, "tps": 664.7666224795696, "wall": 73.93873028201051, "pu": 40.739285714285714, "du": null, "apc": 0.175}, {"name": "reuse_p8192_d2048_o128_2P+6D_rep1", "arm": "2P+6D", "n": 384, "req": 384, "e2e_p50": 1.227345287508797, "e2e_p90": 2.1255404052266385, "e2e_p99": 3.508073750929325, "e2e_mean": 1.4636097900368743, "ttft_p90": 1.4201186655962372, "tpot_p99": 0.006487326852717862, "tps": 513.1391061059077, "wall": 95.78689173195744, "pu": 54.34782608695652, "du": 41.78260869565217, "apc": 0.7}, {"name": "reuse_p8192_d2048_o128_4P+4D_rep1", "arm": "4P+4D", "n": 384, "req": 384, "e2e_p50": 1.0309503950120416, "e2e_p90": 1.806706921494333, "e2e_p99": 2.7563181719661243, "e2e_mean": 1.2135527645268667, "ttft_p90": 1.0876598956412638, "tpot_p99": 0.006993451918086668, "tps": 600.3784995590349, "wall": 81.86835477303248, "pu": 26.28846153846154, "du": 78.27564102564102, "apc": 0.7}, {"name": "reuse_p8192_d2048_o128_6P+2D_rep1", "arm": "6P+2D", "n": 384, "req": 384, "e2e_p50": 1.2513171324972063, "e2e_p90": 1.9010370597010477, "e2e_p99": 2.9169839132187194, "e2e_mean": 1.3710599802317727, "ttft_p90": 0.9651577884971629, "tpot_p99": 0.009227863290451657, "tps": 547.5620167304908, "wall": 89.7651745339972, "pu": 19.189922480620154, "du": 92.96511627906976, "apc": 0.7}, {"name": "reuse_p8192_d2048_o128_8C-proxy_rep1", "arm": "colo", "n": 384, "req": 384, "e2e_p50": 1.3169660340063274, "e2e_p90": 2.2136766588897445, "e2e_p99": 3.0296990163286686, "e2e_mean": 1.4168819271167195, "ttft_p90": 0.8181633891130331, "tpot_p99": 0.019182177199455072, "tps": 509.5195788549742, "wall": 96.46734304196434, "pu": 39.01358695652174, "du": null, "apc": 0.7}]
|