On multi-path bandwidth scheduling for multiple fixed-slot reservations in high-performance networks

Large-scale applications in various fields generate big data at a high speed, which needs to be transferred through High-Performance Networks (HPNs). Bandwidth scheduling in HPNs plays a pivotal role in providing Quality of Service (QoS) to such applications. In this paper, we formulate two bandwidth scheduling problems: (i) maximize the amount of bandwidths to be allocated for multiple fixed-slot bandwidth reservation requests (FBRRs), referred to as maxAB-MultF, and (ii) maximize the number of requests to be satisfied as well as the degree of user's satisfaction for multiple prioritized fixed-slot bandwidth reservation requests (PFBRRs), referred to as maxNR-MultPF. We prove both of these problems to be NP-complete, and design two heuristic algorithms, namely, Maximal Reserved Bandwidth Resources (MaxRBR) and Dynamic Resource Occupation (DynRO), for them, respectively. For each problem, we also design two algorithms based on greedy strategy for performance comparison. Extensive simulation results illustrate that MaxRBR for maxAB-MultF improves Reserved Bandwidth Resource Ratio over two other algorithms in comparison by 18% and 15%, respectively, and DynRO for maxNR-MultPF improves over another two algorithms in comparison scheduling success ratio by 10% and 5%, respectively, and user satisfaction by 6.5% and 5.5%, respectively. Considering the rapid expansion of HPNs in both speed and scale, the proposed scheduling algorithms have great potentials to improve the network performance of big data applications that require these two types of services.