טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentElster Constantine
SubjectEfficient Monitoring of QOS Parameters
DepartmentDepartment of Computer Science
Supervisor Professor Dan Raz


Abstract

With the development of modern Internet applications, such as real time audio and video, the original ``best-effort'' design of the Internet Protocol (IP) is no longer sufficient. This has led in the past few years to the development of the Quality of Service (QoS) concept, in which applications can request, and the network can provide the resources needed to guarantee the required service level. This allows Internet Service Providers (ISPs) to offer predictable service levels in terms of data throughput capacity (bandwidth), latency variations (jitter) and propagation latency. Thus, maintaining updated information regarding these parameters in the flows in order to verify that they satisfy the QoS requirements, as specified in the Service Level Agreements (SLAs), is one of the biggest challenges in the current QoS architecture.

Known oblivious and reactive monitoring techniques do not scale well when the number of flows and the length of their paths increase, and when the network load increases. This is due both to load on the centralized bandwidth allocation entity and to the excessive number of monitoring and control messages needed. In this research we propose a new monitoring technique called Adaptive Monitoring, in which the routers along the flow are responsible to discover and notify the centralized allocation entity when a violation of the SLA occurs (or is soon to occur). We study the performance of this new algorithm through theoretical analysis and extensive simulations. Our results indicate that in addition to dramatically reducing the load from the centralized allocation entity, the amount of network traffic needed is relatively small, and this new monitoring scheme scales well. In fact, even when one third of the links experience higher than expected load, our algorithm performs better than any existing monitoring algorithm.