Dynamic, Distributed, and Optimized: Orchestrating Live Migrating Computations in the Continuum

Emerging Edge applications combine real-time responsiveness with adaptive, data-driven computation that may need to relocate itself to accomplish its assigned goals. Enabling live migration, i.e., resuming as opposed to restarting a running component that needs to relocate, requires portable execution, checkpoint/restore (C/R) capabilities, and ad-hoc orchestration strategies. To this end, we propose a two-tier orchestration framework that separates infrastructure-level concerns from application-level (migratory) ones. Migrating jobs execute as aperiodic requests served by sporadic servers under fixed-priority scheduling, which bounds the interference they impose on node-bound jobs under the assumed scheduling model. Migration placement is formulated as a distributed optimization solved via the Alternating Direction Method of Multipliers (ADMM), minimizing estimated completion time while enforcing resource and locality constraints. Our proof-of-concept, implemented in Rust and based on WebAssembly (Wasm) with compiled-module C/R, allows live migration across heterogeneous Edge nodes with no need for runtime changes. Our initial experiments show that the ADMM solver converges rapidly with low communication overhead and that live migration outperforms its stateless restart-based alternative as well as static execution.