File size: 20,204 Bytes
96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 e83942e 817f4f7 e83942e 817f4f7 e83942e 817f4f7 e83942e 817f4f7 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 e83942e 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 96bb363 b598e06 817f4f7 b598e06 817f4f7 b598e06 817f4f7 b598e06 817f4f7 96bb363 e83942e 96bb363 b598e06 96bb363 e83942e 96bb363 b598e06 96bb363 b598e06 96bb363 2ee9c68 96bb363 817f4f7 96bb363 e83942e 96bb363 e83942e 96bb363 e83942e 96bb363 e83942e 817f4f7 e83942e 817f4f7 e83942e 96bb363 b598e06 96bb363 e83942e 96bb363 e83942e b598e06 | 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 | """
server.py
Runs a simulation between AI datacenter workloads and an electrical grid (IEEE 13-bus OpenDSS model).
Uses GPU power traces and workloads to model howAI inference/training affects grid voltage and stability over time.
"""
from fractions import Fraction
from pathlib import Path
import subprocess, tempfile, os, uvicorn, threading, math, json, hashlib
import pandas as pd
from fastapi import FastAPI, HTTPException, Response
from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
from typing import Optional
from fastapi import WebSocket, WebSocketDisconnect
from openg2g.coordinator import Coordinator
from openg2g.datacenter.config import (
DatacenterConfig,
InferenceModelSpec,
PowerAugmentationConfig,
TrainingRun,
ReplicaSchedule,
)
from openg2g.datacenter.offline import OfflineDatacenter, OfflineWorkload
from openg2g.datacenter.workloads.inference import InferenceData
from openg2g.datacenter.workloads.training import TrainingTrace, TrainingTraceParams
from openg2g.grid.opendss import OpenDSSGrid
from openg2g.grid.config import TapPosition
from openg2g.controller.tap_schedule import TapScheduleController
from openg2g.metrics.voltage import compute_allbus_voltage_stats
import logging
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s %(levelname)s [%(name)s] %(message)s",
)
logger = logging.getLogger(__name__)
import asyncio, time
from concurrent.futures import ProcessPoolExecutor
import json
#currently set to 2 for free tier at hf
_pool = ProcessPoolExecutor(max_workers=2)
_start_time = time.time()
DSS_DIR = Path(__file__).parent / "examples/ieee13"
DSS_MASTER = "IEEE13Nodeckt.dss"
CONFIG_PATH = Path(__file__).parent / "examples/offline/config.json"
# Maps IEEE 13-bus indices to OpenDSS bus names
BUS_INDEX_TO_NAME = {
1:"650", 2:"632", 3:"633", 4:"645", 5:"646", 6:"671",
7:"684", 8:"611", 9:"634", 10:"675", 11:"652", 12:"680", 13:"692",
}
BUSES_ORDERED = [BUS_INDEX_TO_NAME[i] for i in range(1, 14)]
#read files
_config_raw = json.loads(CONFIG_PATH.read_text())
_MODELS = tuple(InferenceModelSpec(**m) for m in _config_raw["models"])
_DC_CONFIG = DatacenterConfig(gpus_per_server=8, base_kw_per_phase=500.0)
if _config_raw.get("data_dir"):
_DATA_DIR = Path(_config_raw["data_dir"])
else:
_DATA_DIR = Path(__file__).parent / "data/specs"
# Load traces_summary.csv once at startup so we can quickly look up trace files
_TRACES_SUMMARY_PATH = _DATA_DIR / "traces_summary.csv"
#Cached dataframe of available GPU power traces
_traces_df: pd.DataFrame | None = None
TAP_STEP = 0.00625
INITIAL_TAPS = TapPosition(
a=1.0 + 14 * TAP_STEP,
b=1.0 + 6 * TAP_STEP,
c=1.0 + 15 * TAP_STEP,
)
# Rescaled to fit in 300s window (original is 3600s, we compress ~12x)
TAP_CHANGE_SCHEDULE = (
TapPosition(
a=1.0 + 16 * TAP_STEP,
b=1.0 + 6 * TAP_STEP,
c=1.0 + 17 * TAP_STEP,
).at(t=75) # was 1500s → 75s at 12x compression
| TapPosition(
a=1.0 + 10 * TAP_STEP,
b=1.0 + 6 * TAP_STEP,
c=1.0 + 10 * TAP_STEP,
).at(t=200) # was 3300s → 200s at 12x compression
)
"""
Load trace index CSV and cache it.
"""
def _load_traces_index() -> pd.DataFrame:
global _traces_df
if _traces_df is None:
if _TRACES_SUMMARY_PATH.exists():
_traces_df = pd.read_csv(_TRACES_SUMMARY_PATH)
else:
_traces_df = pd.DataFrame(columns=["model_label","num_gpus","max_num_seqs","trace_file"])
return _traces_df
"""
Lookup GPU power trace and scale by replica count.
Returns a list of per-timestep total power values in watts.
"""
def _get_trace_power(model_label: str, num_gpus: int, max_num_seqs: int,
num_replicas: int = 1) -> list[float]:
df = _load_traces_index()
row = df[
(df["model_label"] == model_label) &
(df["num_gpus"] == num_gpus) &
(df["max_num_seqs"]== max_num_seqs)
]
if row.empty:
raise ValueError(
f"No trace found for model={model_label} num_gpus={num_gpus} "
f"max_num_seqs={max_num_seqs}. "
f"Available: {df[['model_label','num_gpus','max_num_seqs']].to_dict('records')}"
)
trace_file = _DATA_DIR / row.iloc[0]["trace_file"]
trace_df = pd.read_csv(trace_file)
power_W = trace_df["power_total_W"].tolist()
return [p * num_replicas for p in power_W]
logger.info(f"Data dir: {_DATA_DIR} exists={_DATA_DIR.exists()}")
_load_traces_index() # load at startup
"""Datacenter workload (baseline)"""
def _build_dc(scale: float = 1.0, duration_s: int = 300) -> OfflineDatacenter:
df = _load_traces_index()
first_row = df.iloc[0]
first_model = tuple(m for m in _MODELS if m.model_label == first_row["model_label"])
inference_data = InferenceData.load(_DATA_DIR, first_model)
training_trace = TrainingTrace.ensure(
_DATA_DIR / "training_trace.csv",
TrainingTraceParams(),
)
t0 = min(40.0, duration_s * 0.13)
t1 = min(140.0, duration_s * 0.47)
replica_schedules = {}
for m in _MODELS:
initial_replicas = max(1, int(scale * 8))
reduced_replicas = max(1, int(initial_replicas * 0.25))
replica_schedules[m.model_label] = (
ReplicaSchedule(initial=initial_replicas)
.ramp_to(
reduced_replicas,
t_start=min(150.0, duration_s * 0.50),
t_end=min(220.0, duration_s * 0.73),
)
)
workload = OfflineWorkload(
inference_data=inference_data,
replica_schedules=replica_schedules,
training=TrainingRun(
n_gpus=max(1, int(24 * scale)),
trace=training_trace,
target_peak_W_per_gpu=400.0,
).at(
t_start=t0,
t_end=t1,
),
)
return OfflineDatacenter(
_DC_CONFIG,
workload,
dt_s=Fraction(1, 10),
seed=0,
name="baseline",
total_gpu_capacity=1000,
power_augmentation=PowerAugmentationConfig(
amplitude_scale_range=(0.88, 1.12),
noise_fraction=0.04,
),
)
"""
Build datacenter workload from GPU trace.
Returns (datacenter, raw_power_W_list)
"""
def _build_dc_from_real_trace(
model_label: str,
num_gpus: int,
max_num_seqs: int,
num_replicas: int,
duration_s: int,
) -> tuple[OfflineDatacenter, list[float]]:
power_W = _get_trace_power(model_label, num_gpus, max_num_seqs, num_replicas)
target_steps = int(duration_s / 0.1)
if len(power_W) < target_steps:
repeats = math.ceil(target_steps / len(power_W))
power_W = (power_W * repeats)[:target_steps]
else:
power_W = power_W[:target_steps]
df = _load_traces_index()
row = df[df["model_label"] == model_label].iloc[0]
model_tuple = tuple(m for m in _MODELS if m.model_label == model_label)
inference_data = InferenceData.load(_DATA_DIR, model_tuple)
workload = OfflineWorkload(
inference_data=inference_data,
replica_schedules={
model_label: ReplicaSchedule(initial=num_replicas)
},
)
# ← compute actual GPU count and add headroom
actual_gpu_count = num_replicas * num_gpus
gpu_capacity = max(1000, actual_gpu_count * 2)
dc = OfflineDatacenter(
_DC_CONFIG, workload, dt_s=Fraction(1, 10), seed=0,
name=model_label.replace(".", "-"),
total_gpu_capacity=gpu_capacity, # ← was hardcoded 1000
power_augmentation=PowerAugmentationConfig(
amplitude_scale_range=(1.0, 1.0),
noise_fraction=0.0,
),
)
return dc, power_W
"""Create IEEE 13-bus grid with datacenter connection."""
def _build_grid(tap_pu: float, dc_bus: str) -> OpenDSSGrid:
return OpenDSSGrid(
dss_case_dir=str(DSS_DIR),
dss_master_file=DSS_MASTER,
dt_s=Fraction(1),
source_pu=tap_pu,
initial_tap_position=INITIAL_TAPS,
)
def _make_tap(v: float):
return TapPosition(a=v, b=v, c=v).at(t=0)
"""Run datacenter + grid simulation."""
def _run(dc, grid, tap_pu, dc_bus, duration_s):
grid.attach_dc(dc, bus=dc_bus, connection_type="wye", power_factor=_DC_CONFIG.power_factor)
coord = Coordinator(
datacenters=[dc],
grid=grid,
controllers=[TapScheduleController(
schedule=TAP_CHANGE_SCHEDULE, # ← real schedule
dt_s=Fraction(1)
)],
total_duration_s=duration_s,
)
return coord.run()
"""
Runs one full simulation job (datacenter + grid) in a worker process
and returns results for the API.
"""
def _run_full(req_dict: dict) -> dict:
dc_bus = BUS_INDEX_TO_NAME.get(req_dict["targetBus"], "671")
replicas = max(1, req_dict["numReplicas"])
dc, raw_power_W = _build_dc_from_real_trace(
model_label = req_dict["modelLabel"],
num_gpus = req_dict["numGpus"],
max_num_seqs = req_dict["maxNumSeqs"],
num_replicas = replicas,
duration_s = req_dict["durationS"],
)
grid = _build_grid(req_dict["substationVoltage"], dc_bus)
log = _run(dc, grid, req_dict["substationVoltage"], dc_bus, req_dict["durationS"])
step = max(1, req_dict["sampleInterval"])
gs_sampled = log.grid_states[::step]
t_sampled = list(log.time_s[::step])
dc_states = log.dc_states
results = []
for i, (t, gs) in enumerate(zip(t_sampled, gs_sampled)):
vs = _voltages(gs)
dc_i = min(range(len(dc_states)), key=lambda j: abs(dc_states[j].time_s - t))
ds = dc_states[dc_i]
kw = float((ds.power_w.a + ds.power_w.b + ds.power_w.c) / 1000)
if math.isnan(kw): kw = 0.0
trace_idx = min(int(t / 0.1), len(raw_power_W) - 1) if raw_power_W else 0
raw_kw = raw_power_W[trace_idx] / 1000.0 if raw_power_W else kw
results.append({
"time": float(t),
"gpu_power_W": kw * 1000,
"gpu_power_kW": kw,
"gpu_power_raw_kW": raw_kw,
"gpu_reactive_kVAR": kw * 0.329,
"active_gpus": replicas * req_dict["numGpus"],
"voltages": vs,
"min_voltage": min(vs),
"max_voltage": max(vs),
"target_bus_voltage": vs[req_dict["targetBus"] - 1],
"total_load_kW": kw,
})
return {
"numSamples": len(results),
"targetBus": req_dict["targetBus"],
"modelLabel": req_dict["modelLabel"],
"numGpus": req_dict["numGpus"],
"maxNumSeqs": req_dict["maxNumSeqs"],
"numReplicas": replicas,
"duration": float(max(r["time"] for r in results) if results else 0),
"minVoltage": float(min(r["min_voltage"] for r in results) if results else 1.0),
"maxVoltage": float(max(r["max_voltage"] for r in results) if results else 1.0),
"avgGpuPower": float(sum(r["gpu_power_W"] for r in results) / len(results) if results else 0),
"peakGpuPower": float(max(r["gpu_power_W"] for r in results) if results else 0),
"timeSeries": results,
}
"""Get per-bus voltage (worst phase per bus)."""
def _voltages(gs) -> list[float]:
result = []
none_count = 0
for name in BUSES_ORDERED:
try:
tp = gs.voltages[name]
vals = [float(v) for v in [tp.a, tp.b, tp.c]
if not math.isnan(float(v)) and 0.5 < float(v) < 1.5]
result.append(min(vals) if vals else None)
except Exception as e:
logger.debug(f"Bus {name} voltage unavailable: {e}")
result.append(None)
none_count = sum(1 for v in result if v is None)
if none_count > 3:
logger.warning(f"OpenDSS convergence failure: {none_count}/13 buses returned None")
known = [v for v in result if v is not None]
avg = sum(known) / len(known) if known else 1.0
return [v if v is not None else avg for v in result]
# ── FastAPI────────────────────────────────────────────────────────────────
app = FastAPI()
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=False,
allow_methods=["*"],
allow_headers=["*"],
)
class PowerflowRequest(BaseModel):
substationVoltage: float = 1.05
numBuses: int = 13
baseVoltage: float = 4.16
targetBus: int = 0
class LLMImpactRequest(BaseModel):
targetBus: int = 9
sampleInterval: int = 1
substationVoltage: float = 1.05
modelLabel: str = "Llama-3.1-8B"
numGpus: int = 1
maxNumSeqs: int = 128
numReplicas: int = 1
durationS: int = 300
class HeatmapRequest(BaseModel):
voltages: list[float]
dataCenterBus: Optional[int] = None
@app.get("/api/health")
def health():
return {"status": "ok", "data_ready": _DATA_DIR.exists(),
"message": "gpu2grid OpenDSS server"}
"""Return available traces"""
@app.get("/api/traces")
def list_traces():
df = _load_traces_index()
if df.empty:
return {"traces": [], "models": [], "trainingAvailable": False}
traces = df[["model_label","num_gpus","max_num_seqs"]].to_dict("records")
models = []
for model_label, group in df.groupby("model_label"):
models.append({
"modelLabel": model_label,
"numGpus": int(group["num_gpus"].iloc[0]),
"batchSizes": sorted(group["max_num_seqs"].tolist()),
})
training_available = (_DATA_DIR / "training_trace.csv").exists()
return {
"traces": traces,
"models": models,
"trainingAvailable": training_available,
"dataDir": str(_DATA_DIR),
}
"""Baseline grid simulation, no workload"""
@app.post("/api/powerflow")
async def powerflow(req: PowerflowRequest):
logger.info(f"Powerflow request v={req.substationVoltage}")
try:
dc = _build_dc(scale=0.001, duration_s=5)
grid = _build_grid(req.substationVoltage, "671")
log = _run(dc, grid, req.substationVoltage, "671", 5)
vs = _voltages(log.grid_states[-1])
logger.info(f"Powerflow result min={min(vs):.4f} max={max(vs):.4f}")
return {"buses": [{"id": i+1, "voltage": v, "activePower": 0.0,
"reactivePower": 0.0} for i, v in enumerate(vs)],
"lines": []}
except Exception as e:
logger.exception("Powerflow failed")
raise HTTPException(status_code=500, detail=str(e))
"""Simulate AI workload impact on grid using GPU traces."""
@app.post("/api/llm-impact")
async def llm_impact(req: LLMImpactRequest):
logger.info(f"Simulation request: {req.modelLabel} bus={req.targetBus}")
try:
loop = asyncio.get_event_loop()
result = await loop.run_in_executor(_pool, _run_full, req.dict())
return result
except Exception as e:
logger.exception("Simulation failed")
raise HTTPException(status_code=500, detail=str(e))
@app.post("/api/heatmap")
async def heatmap(req: HeatmapRequest):
if len(req.voltages) != 13:
raise HTTPException(400, f"Need 13 voltages, got {len(req.voltages)}")
script = str(Path(__file__).parent / "generate_heatmap.py")
with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as f:
out = f.name
subprocess.run(
["python3", script, out] + [str(v) for v in req.voltages] +
([str(req.dataCenterBus)] if req.dataCenterBus else []),
check=True,
)
png = open(out, "rb").read()
os.unlink(out)
return Response(content=png, media_type="image/png")
def _serialize_tick(tick, req_dict: dict, raw_power_W: list[float]) -> dict:
"""Serialize one TickOutput to a small JSON-safe dict for the frontend."""
t = tick.t_s
# Grid voltages (may be None if grid didn't tick this step)
voltages = None
min_v = max_v = target_v = None
if tick.grid_state is not None:
voltages = _voltages(tick.grid_state)
min_v = min(voltages)
max_v = max(voltages)
target_v = voltages[req_dict["targetBus"] - 1]
# DC power (sum all sites)
kw = 0.0
batch_by_model: dict[str, int] = {}
for dc_name, ds in tick.dc_states.items():
pw = ds.power_w
kw += float((pw.a + pw.b + pw.c) / 1000)
if hasattr(ds, "batch_size_by_model"):
batch_by_model.update(ds.batch_size_by_model)
if math.isnan(kw):
kw = 0.0
trace_idx = min(int(t / 0.1), len(raw_power_W) - 1) if raw_power_W else 0
raw_kw = raw_power_W[trace_idx] / 1000.0 if raw_power_W else kw
events = [{"type": e.event_type, "data": e.data} for e in tick.sim_events]
return {
"time": float(t),
"gpu_power_kW": kw,
"gpu_power_raw_kW": raw_kw,
"active_gpus": req_dict["numReplicas"] * req_dict["numGpus"],
"batch_by_model": batch_by_model,
"voltages": voltages,
"min_voltage": min_v,
"max_voltage": max_v,
"target_bus_voltage": target_v,
"sim_events": events,
}
def _run_streaming(req_dict: dict):
"""Generator: builds coordinator and yields serialized tick dicts."""
dc_bus = BUS_INDEX_TO_NAME.get(req_dict["targetBus"], "671")
replicas = max(1, req_dict["numReplicas"])
dc, raw_power_W = _build_dc_from_real_trace(
model_label = req_dict["modelLabel"],
num_gpus = req_dict["numGpus"],
max_num_seqs = req_dict["maxNumSeqs"],
num_replicas = replicas,
duration_s = req_dict["durationS"],
)
grid = _build_grid(req_dict["substationVoltage"], dc_bus)
grid.attach_dc(dc, bus=dc_bus, connection_type="wye",
power_factor=_DC_CONFIG.power_factor)
coord = Coordinator(
datacenters=[dc],
grid=grid,
controllers=[TapScheduleController(
schedule=TAP_CHANGE_SCHEDULE,
dt_s=Fraction(1),
)],
total_duration_s=req_dict["durationS"],
)
step = max(1, req_dict["sampleInterval"])
tick_num = 0
try:
for tick in coord.run_iter():
if tick_num % step == 0:
yield _serialize_tick(tick, req_dict, raw_power_W)
tick_num += 1
finally:
coord.stop()
@app.websocket("/ws/sim-stream")
async def sim_stream(ws: WebSocket):
await ws.accept()
try:
req_dict = await ws.receive_json()
req = LLMImpactRequest(**req_dict)
logger.info(f"WS stream: {req.modelLabel} bus={req.targetBus}")
# Run full simulation in process pool (separate process = safe for OpenDSS)
loop = asyncio.get_event_loop()
result = await loop.run_in_executor(_pool, _run_full, req.dict())
# Stream results tick by tick from the completed result
for row in result["timeSeries"]:
await ws.send_json(row)
await ws.send_json({"done": True})
except WebSocketDisconnect:
logger.info("WS client disconnected")
except Exception as e:
logger.exception("WS stream failed")
try:
await ws.send_json({"error": str(e)})
except Exception:
pass
if __name__ == "__main__":
logger.info("=" * 70)
logger.info(f"Data dir: {_DATA_DIR} ready={_DATA_DIR.exists()}")
df = _load_traces_index()
if not df.empty:
models = df["model_label"].unique().tolist()
logger.info(f"Models: {models}")
logger.info(f"Traces: {len(df)} configurations")
logger.info("=" * 70)
uvicorn.run("server:app", host="0.0.0.0", port=8080, workers=1, log_level="info", ws_ping_interval=None)
|