| We address a problem of data packet scheduling in high-speed computer networks. We formally define requirements for a high-speed packet scheduler and argue that the existing algorithms either are impractical or do not provide the full spectrum of required features. We consider a new approach, called hybrid scheduling, that is capable of providing all the required features, but we argue that existing hybrid algorithms do not provide them. We then propose a hybrid scheduling algorithm that satisfies all the required properties and we develop a theoretical framework to prove that. In doing so, we extend an existing theoretical framework to a more general case and also resolve several misconceptions that are present in recent publications on the subject. We describe the design of a real system that implements a hybrid scheduler on a parallel network processing architecture. We demonstrate that our design approach results in a flexible and efficient implementation. Finally, we present experimental results obtained on the prototype system as well as on simulations, suggesting that a hybrid scheduling approach, if properly designed and developed, can provide desirable scheduling characteristics in high-speed networks. |
