Slack allocation techniques for intra-path load balancing
Journal of High Speed Networks
Novel dynamic delay allocation adjustment for improving bandwidth efficiency
Computer Communications
| Hi-index | 0.00 |
In recent years, large commercial and non-commercial customers of network service providers (NSP) are increasingly demanding dedicated and fault-tolerant wide-area network connectivity between their remote endpoints with the ability to exercise fine-grained control over their share of physical network resources. This trend is being coupled with a rapid surge in the amount of real-time network traffic that the NSP's network infrastructure is required to carry. In other words, customers require a virtualized share of the NSP's physical network infrastructure with performance and isolation guarantees. At the same time, NSPs themselves have an inherent need to maximize their own revenue base by accommodating the requirements of as many customers as possible. These three competing forces have created an urgent requirement for resource provisioning techniques that enable network resource virtualization with performance and isolation guarantees and at the same time maximize the utilization efficiency of the network infrastructure. This dissertation presents a comprehensive set of resource provisioning techniques to satisfy these competing requirements on the network infrastructure that carries traffic with performance constraints, also popularly known as Quality of Service (QoS) constraints. We propose three-levels of inter-dependent resource provisioning algorithms for real-time traffic flows that require long-term bandwidth guarantees and deterministic or statistical end-to-end delay bounds. At the link-level, we have developed a novel measurement-based algorithm called Delay Distribution Measurement (DDM) based admission control, that provides distinct per-flow statistical delay guarantees. DDM can increase link utilization for voice traffic by up to a factor of 3 compared to deterministic admission control when tolerance to delay violation is as low as 10−5. At the path-level, we have developed an algorithm called Load-based Slack Sharing (LSS), that partitions the end-to-end QoS requirement of a flow into QoS requirements at each hop along the path. The goal of partitioning is to keep the loads on different hops as balanced as possible. Compared to best previously known approach, LSS can admit up to 1.2 times more traffic with deterministic delay bounds and up to 2.8 times with statistical delay bounds. At the network-level, we propose a route-selection algorithm called Link Criticality Based Routing (LCBR). LCBR algorithm selects primary and backup routes between a given source and destination under end-to-end delay and bandwidth constraints. It applies a simple notion of link importance to select primary and backup routes that satisfy the end-to-end QoS while balancing the load across the network. LCBR can improve the amount of admitted network traffic by up to a factor of 2 compared to its variants that use route selection criteria from earlier traffic engineering approaches.