Edge Server Handoff: Mobility-Aware Migration Simulator (2D)

Device Path

Device speed 1.0×

Handoff Policy

Hysteresis margin 15 ms

Migration cost 120 ms

Edge nodes 6

Live Session (smart policy)

Serving node RTT

— ms

Effective latency

— ms

Handoffs so far

Ping-pong (<3s apart)

Mean latency

— ms

Time in migration

vs. Naive "always nearest"

Smart total cost

Naive total cost

Naive handoffs

Smart saves

How it works

Each edge node i sits at a fixed 2D point on the map; the mobile device moves along a route. Round-trip time to a node is a propagation + processing model driven by real on-screen distance:

RTT_i = 2 · dist(device, node_i) / v_prop + T_proc

v_prop ≈ 2×10^8 m/s (fibre backhaul)
T_proc = fixed per-node processing delay (4–12 ms)

The smart policy stays attached to its current serving node even when a closer node exists, because switching costs time. A handoff only fires when a candidate is cheaper by more than the hysteresis margin — the same anti-flapping rule real cellular and Wi-Fi handover uses:

handoff  ⇔  RTT_candidate + margin  <  RTT_serving

During a handoff the session state (containers, cached model weights, TCP context) must migrate to the new node, which stalls effective latency by the configured migration cost for a short window:

latency_effective(t) = RTT_serving(t)                      normally
                     = RTT_serving(t) + migration_cost      during an active migration

A naive shadow policy runs the same device path in parallel and always migrates to whichever node is instantaneously nearest (margin = 0), paying the same migration cost on every switch. The two "total cost" readouts accumulate effective latency over the identical run, so the difference is a genuine, measured comparison of the two policies:

Margin = 0 on the smart policy makes it behave like the naive one — watch its ping-pong counter climb near node boundaries.
Margin too high keeps a stale, distant node long after a much closer one is available, inflating mean latency even though handoffs are rare.
Switch to the zig-zag route, which repeatedly grazes a boundary between two nodes — that's where hysteresis earns its keep: the naive policy racks up ping-pong handoffs it doesn't need, while the smart policy suppresses most of them and comes out ahead on total cost. On the smooth loop route there is little boundary-hugging, so raising the migration cost slider is what tips the balance in the smart policy's favor there.

This is the exact trade-off mobile edge computing platforms (AWS Wavelength, Azure Edge Zones, 5G MEC) solve when deciding whether to migrate a live AR/VR or inference session to a newly-nearer edge server as a user moves.

Serving edge node (smart) Idle edge node Naive-policy target Mobile device

Serving: node 0

State: connected