Performing accurate simulations for deadline-aware applications

Guthemberg Silvestre (1) , Sébastien Monnet (1)

1 Regal - Large-Scale Distributed Systems and Applications

One of the most appealing aspects of cloud computing is its potential capacity of offering customized services for clients through Service Level Agreement (SLA) contracts. Emerging cloud providers are intended to provide services whose performance targets are defined by precise data transfer deadlines. Yet, enforcing strict transfer rates for cloud-like applications is not simple. This has caught system designers and analysts' interest, and drawn their attention to study and evaluate the performance of deadline-aware applications. In order to address the increasing demand for performance evaluation tools for deadline-aware simulations, we present a bandwidth scheduling component for the PeerSim simulator. Our component provides a lightweight and accurate fair-sharing of bandwidth and rate control enforcement. We assume a simple approach that does not reproduce in-depth transport protocol behaviour. Instead, our design focuses in reducing the inherent impact on the overall simulation scalability as bandwidth scheduling accuracy is improved. We have studied the scalability and bandwidth scheduling accuracy of four scheduling approaches, namely the simplest, lock-based, packet or slot-based, and connection-oriented bandwidth scheduler. We have verified that a connection-oriented approach allows us to (i) improve considerably the accuracy of fair bandwidth sharing, and (ii) implement a rate control mechanism properly. Our source code is available online.

Domaines: Réseaux et télécommunications [cs.NI]

Guthemberg Silvestre, Sébastien Monnet. Performing accurate simulations for deadline-aware applications. HPCS 2013 - The 2013 International Conference on High Performance Computing & Simulation, Jul 2013, Helsinki, Finland. pp.65-71, ⟨10.1109/HPCSim.2013.6641394⟩. ⟨hal-00861970⟩