Construction and Evaluation of a Service Level Agreement Test-Bed

Abstract

Differentiated Services (DiffServ) enable quality-of-service provisioning by coupling traffic aggregates with a specific forwarding path treatment. Per-Hop forwarding Behavior (or PHB) describes the forwarding path behavior required in network nodes, while the Per-Domain Behavior (or PDB) describes the behavior experienced by a particular set of packets as they cross a DiffServ domain. PHB and PDB implementations with traffic conditioners, provisioning strategies and billing models are building blocks used in Services Level Agreements (SLA).The goal of the project was to construct an SLA test-bed using soft DiffServ routing nodes. This test-bed, and the associated tools, is intended for in-depth studies of SLA translation and mapping algorithms. The specific objectives of the work reported in this thesis were to construct the test-bed, and then verify, validate and evaluate different elements of the SLA test-bed. Specifically, proof-of-concept experiments were conducted to show that one can implement a general Services Level Agreement (SLA) based on DiffServ, to show how to translate SLA requirements into real test-bed parameters, and then run experiments, collect data and make performance measurements.. The study is empirical and involves implementation of some of the quality of service (QoS) related IETF drafts and RFCs in a five-node soft DiffServ-based SLA testbed. This includes implementation and empirical evaluation of Expedite Forwarding (EF) PHB, Assured Forwarding (AF) PHB, Virtual Wire (VW) PDB, and of Assured Forwarding PDB.Virtual Wire PDB was implemented using both the Priority Scheduler and a Weighted Fair Queuing Scheduler. The effect of Best Effort packet size and rate on the protected EF traffic was investigated empirically. This provided proof of concept data on how successful can a VW implementation be in protecting Voice over IP (VoIP) streams. We used both emulated VoIP streams, and real VoIP phones in highly congested traffic environments.Assured Rate PDB was implemented using appropriate traffic conditioning, buffer management and scheduling techniques. The issue of what to do with non-conforming AF traffic was investigated empirically - whether to drop the non-conforming packets or to downgrade them to Best Effort. It was noticed that strict dropping of the non-conforming packets will sometimes underutilize the channel, while downgrading them to Best Effort will create a reordering problem.