Edge Offloading: Energy vs Latency Simulator (2D)

Task Parameters

Task data size 120 KB

Local CPU frequency 1.2 GHz

Wireless bandwidth 10 Mbps

Policy weight α (energy ⟷ latency) 0.50

Decision Policy

Boundary Map Axes

Predicted per Task

Local time

0 ms

Offload time

0 ms

Local energy

0 mJ

Offload energy

0 mJ

Live Session Stats

Battery used

0.0 mJ

Offload ratio

Tasks completed

Tasks in flight

How it works

Each task of size D needs C = D × L processor cycles, where L is a fixed workload density (cycles per bit — a stand-in for how compute-heavy the task is). Two routes compete for every task:

Local:    T_local = C / f_local
          E_local = C × κ × f_local²   (dynamic CMOS power ∝ f²)

Offload:  T_tx      = D / R                (R = wireless bandwidth)
          T_offload = T_tx + C / f_edge     (f_edge is a fast, fixed edge server)
          E_offload = P_tx × T_tx           (radio energy only — the edge server pays its own compute energy)

Adaptive score:  score = α·(T / T_max) + (1-α)·(E / E_max)
                 route with the LOWER score wins this task

Task size / CPU frequency / bandwidth — recompute T and E for both routes live; raising local frequency cuts local time but raises local energy (the f² term), which is the classic mobile-edge-computing trade-off.
α slider — 0 optimizes purely for battery life, 1 optimizes purely for speed, in between blends both (only used in Adaptive mode).
Boundary map — sweeps the exact same score formula across a grid of two parameters (holding the others at their current slider values) and colors each cell by which route actually wins there, so the crossover curve is computed, not drawn.
Always Local / Always Offload / Adaptive — force a route or let the scored policy decide per task, exactly as a real MEC (mobile edge computing) offloading controller would.

This is the standard computation-offloading model used in mobile/IoT edge-computing research (e.g. Kumar & Lu's energy-latency offloading analysis): a device with limited battery deciding, task by task, whether local silicon or a nearby edge server is the better place to run the work. This 2D build renders the identical equations as a bar-chart comparison, an animated packet timeline and a swept decision-boundary map instead of a 3D scene.

Local compute Offloaded task Wireless link