Keywords
software, simulation, networking, computing

Technologies
ns-3, C++, Python

Evolving applications, such as augmented reality and Industry 5 rely on demanding workloads. These workloads are typically offloaded to distributed computing continuum tiers to meet their strict latency requirements. Scalable performance evaluation of these systems is beyond the reach of analysis and experimentation, leaving simulations as the key approach to test the performance of these applications. However, existing simulation tools focus on either the network dynamics, computing operation, or lack features required by complex workloads and advanced computing platforms.

ns3-Continuum extends ns3, a widely used network simulator, to enable simulating distributed computing systems supporting advanced workloads, task offloading and state-of-the-art accelerators. Specifically, ns3-Continuum models offloaded tasks as directed acyclic graphs that capture the dependency of processing pipelines. Additionally, ns3-Continuum GPU accelerators support vertical scaling that can be leveraged for energy-efficient computing resource scheduling. The implementation is designed to facilitate extending end-device behavior, accelerator operation, and control plane decision engines.