File size: 9,707 Bytes
5a22808
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.

"""
Data models for the Thermal Grid Rl Agent Environment.

The agent interacts with a high-dimensional datacenter + power-grid environment.
It observes thermal state, IT load, and external grid signals, then controls:
  A. Cooling (continuous) — CRAC setpoints, fan speeds (VFD), chiller stacking
  B. Load distribution (discrete/structural) — geographical load balancing (GLB)
     and batch-job scheduling
"""

from openenv.core.env_server.types import Action, Observation
from pydantic import Field


# ---------------------------------------------------------------------------
# Action
# ---------------------------------------------------------------------------

class ThermalGridRlAgentAction(Action):
    """
    Control action for the Thermal Grid RL Agent.

    Two distinct levers that interact: redistributing load changes the thermal
    profile, which in turn requires adjustments to cooling.

    A. Cooling Control (Continuous)
    --------------------------------
    crac_setpoint_c   : CRAC/CRAH supply-air temperature setpoint (°C).
    fan_speeds_pct    : Per-rack Variable Frequency Drive fan speed 0–100 %.
                        Power scales as Speed³ (Affinity Laws), so small
                        reductions yield large savings.
    num_active_chillers: How many chiller units to run. A chiller at 20 %
                        load is far less efficient than one at 80 % load.

    B. Load Distribution (Discrete / Structural)
    ---------------------------------------------
    region_traffic_weights : Fraction of incoming user traffic routed to each
                             region (must sum to 1.0). Enables GLB — shift
                             traffic away from hot/expensive regions.
    batch_job_schedule     : Per-batch-job flag — 1 = run now, 0 = defer to
                             off-peak. Enables temporal load shifting.
    workload_matrix        : Per-server utilisation after placement (0–1),
                             shape [num_racks][servers_per_rack].
    power_caps_w           : Per-server power cap in Watts.
    """

    # ---- A. Cooling (continuous) -----------------------------------------
    crac_setpoint_c: float = Field(
        default=18.0,
        description="CRAC/CRAH supply-air temperature setpoint in °C (12.0–27.0)",
    )
    fan_speeds_pct: list[float] = Field(
        default_factory=lambda: [60.0] * 10,
        description=(
            "Per-rack fan speed (0–100 %). Power ∝ Speed³ (Affinity Laws). "
            "One value per rack."
        ),
    )
    num_active_chillers: int = Field(
        default=2,
        description=(
            "Number of chiller units to activate (1–max_chillers). "
            "Fewer chillers at higher load is more efficient than many at low load."
        ),
    )

    # ---- B. Load distribution (discrete / structural) --------------------
    region_traffic_weights: list[float] = Field(
        default_factory=lambda: [1.0],
        description=(
            "Fraction of live user traffic routed to each region (sums to 1.0). "
            "Geographical Load Balancing (GLB) lever."
        ),
    )
    batch_job_schedule: list[int] = Field(
        default_factory=list,
        description=(
            "Per-batch-job scheduling flag: 1 = run this step, 0 = defer to "
            "off-peak. Enables temporal load shifting."
        ),
    )
    workload_matrix: list[list[float]] = Field(
        default_factory=lambda: [[0.5] * 8] * 10,
        description=(
            "Per-server utilisation after placement decisions (0–1). "
            "Shape [num_racks][servers_per_rack]."
        ),
    )
    power_caps_w: list[list[float]] = Field(
        default_factory=lambda: [[300.0] * 8] * 10,
        description="Per-server power cap in Watts. Shape [num_racks][servers_per_rack].",
    )
    metadata: dict = Field(default_factory=dict, description="Metadata dictionary for tracking action provenance")


# ---------------------------------------------------------------------------
# Observation
# ---------------------------------------------------------------------------

class ThermalGridRlAgentObservation(Observation):
    """
    Observation returned by the Thermal Grid RL Agent environment.

    Groups into four sections mirroring the environment's key components:
      1. Thermal state   — rack inlet/CPU temps, thermal mass lag indicator
      2. IT load         — per-rack power, utilisation, live vs batch split
      3. Cooling state   — PUE, CRAC/chiller status, fan speeds
      4. External grid   — energy price, carbon intensity, demand-response flag
    """

    step_summary: str = Field(default="", description="Step summary string")

    # ---- 1. Thermal state ------------------------------------------------
    inlet_temps_c: list[float] = Field(
        default_factory=list,
        description="Per-rack cold-aisle inlet temperature in °C",
    )
    mean_cpu_temps_c: list[float] = Field(
        default_factory=list,
        description="Per-rack mean CPU junction temperature in °C",
    )
    max_cpu_temps_c: list[float] = Field(
        default_factory=list,
        description="Per-rack maximum CPU junction temperature in °C",
    )
    max_gpu_temps_c: list[float] = Field(
        default_factory=list,
        description="Per-rack maximum GPU junction temperature in °C",
    )
    thermal_mass_lag_c_per_min: float = Field(
        default=0.0,
        description=(
            "Estimated rate of temperature change (°C/min) due to thermal mass. "
            "Positive = room heating up; negative = cooling down. "
            "Enables predictive rather than purely reactive control."
        ),
    )

    # ---- 2. IT load ------------------------------------------------------
    rack_powers_w: list[float] = Field(
        default_factory=list,
        description="Per-rack total IT power draw in Watts",
    )
    rack_utilisation: list[float] = Field(
        default_factory=list,
        description="Per-rack mean server utilisation (0–1)",
    )
    live_traffic_load_w: float = Field(
        default=0.0,
        description="Power attributable to live user-facing traffic (Watts)",
    )
    deferred_batch_load_w: float = Field(
        default=0.0,
        description="Power attributable to batch jobs running this step (Watts)",
    )
    pending_batch_jobs: int = Field(
        default=0,
        description="Number of batch jobs currently queued / deferred",
    )

    # ---- 3. Cooling state ------------------------------------------------
    pue: float = Field(
        default=0.0,
        description="Power Usage Effectiveness = total facility power / IT power",
    )
    total_it_power_w: float = Field(default=0.0, description="Total IT power draw in Watts")
    # ---- Episode metadata ------------------------------------------------
    done: bool = Field(default=False, description="1 if the episode has ended")
    reward: float = Field(default=0.0, description="Reward from the environment")
    safety_override_triggered: bool = Field(
        default=False,
        description="True if the Hard Safety Layer overrode agent actions to prevent damage",
    )
    metadata: dict = Field(default_factory=dict, description="Metadata dictionary")
    total_facility_power_w: float = Field(default=0.0, description="Total facility power in Watts")
    crac_power_w: float = Field(default=0.0, description="CRAC/CRAH unit power draw in Watts")
    chiller_power_w: float = Field(default=0.0, description="Active chiller unit power draw in Watts")
    num_active_chillers: int = Field(
        default=0,
        description="Number of chiller units currently running",
    )
    chiller_load_pct: float = Field(
        default=0.0,
        description=(
            "Mean load fraction across active chillers (0–100 %). "
            "Low values (< 40 %) indicate inefficient part-load operation."
        ),
    )
    crac_supply_temp_c: float = Field(
        default=0.0,
        description="Current CRAC supply-air temperature setpoint in °C",
    )
    avg_fan_speed_pct: float = Field(
        default=0.0,
        description="Average VFD fan speed across all racks (0–100 %)",
    )

    # ---- 4. External grid ------------------------------------------------
    ambient_temp_c: float = Field(
        default=25.0,
        description="Current ambient air temperature in °C. Affects chiller and CRAC efficiency.",
    )
    energy_price_per_kwh: float = Field(
        default=0.0,
        description=(
            "Current electricity price ($/kWh). Changes hourly (TOU) or every "
            "5 minutes (Real-Time Pricing). Key signal for temporal load shifting."
        ),
    )
    grid_carbon_intensity_g_per_kwh: float = Field(
        default=0.0,
        description=(
            "Grid carbon intensity (gCO₂/kWh). Low during solar/wind peaks, "
            "high during fossil-fuel peaks."
        ),
    )
    demand_response_signal: int = Field(
        default=0,
        description=(
            "Binary grid demand-response request: 1 = grid requests load shed "
            "to prevent instability, 0 = normal operation."
        ),
    )
    off_peak_window: int = Field(
        default=0,
        description=(
            "1 if current time falls within off-peak pricing window "
            "(e.g. 02:00–05:00), 0 otherwise. Used for batch job scheduling."
        ),
    )