https://joulie-k8s.github.io/Joulie/versions/v0.0.2/docs/simulator/Recent content in Simulator on JoulieHugoen-ushttps://joulie-k8s.github.io/Joulie/versions/v0.0.2/docs/simulator/simulator/Mon, 01 Jan 0001 00:00:00 +0000https://joulie-k8s.github.io/Joulie/versions/v0.0.2/docs/simulator/simulator/<p>This document defines the Joulie simulator design and how it integrates with Joulie.</p> <h2 id="goals">Goals</h2> <ul> <li>Keep Kubernetes scheduling real (real pod placement/lifecycle).</li> <li>Simulate hardware telemetry and control interfaces (RAPL/DVFS now, GPU later).</li> <li>Provide reproducible, comparable experiments across Joulie and WAO.</li> </ul> <h2 id="design-choice-hybrid-simulation">Design choice: hybrid simulation</h2> <p>The simulator is not a fake scheduler.</p> <ul> <li>Pod placement and pod lifetime stay real in Kubernetes.</li> <li>Simulator reads cluster state (pods/nodes) and derives synthetic hardware state per node.</li> <li>Joulie reads telemetry and sends control intents through HTTP endpoints.</li> </ul> <p>This gives one source of truth for workload location: Kubernetes API.</p>https://joulie-k8s.github.io/Joulie/versions/v0.0.2/docs/simulator/simulator-algorithms/Mon, 01 Jan 0001 00:00:00 +0000https://joulie-k8s.github.io/Joulie/versions/v0.0.2/docs/simulator/simulator-algorithms/<p>This page documents the core simulator math and control/workload update loop implemented in <code>simulator/cmd/simulator/main.go</code>.</p> <h2 id="per-node-state">Per-Node State</h2> <p>Each simulated node tracks:</p> <ul> <li><code>CapWatts</code></li> <li><code>TargetThrottlePct</code></li> <li><code>ThrottlePct</code></li> <li><code>FreqScale</code> in <code>[f_min/f_max, 1]</code></li> <li><code>CPUUtil</code> in <code>[0,1]</code></li> <li><code>CapSaturated</code> flag</li> </ul> <h2 id="power-model">Power Model</h2> <p>At telemetry/control update time, simulator computes power:</p> <p><code>P = P_idle + (P_max - P_idle) * util^alpha * freqScale^beta</code></p> <p>Where:</p> <ul> <li><code>P_idle = BaseIdleW</code></li> <li><code>P_max = PMaxW</code></li> <li><code>util = clamp(CPUUtil, 0, 1)</code></li> <li><code>alpha = AlphaUtil</code> (defaults to 1 if invalid)</li> <li><code>beta = BetaFreq</code> (defaults to 1 if invalid)</li> </ul> <p>Then applies bounds/cap handling:</p>