Graduation Semester and Year




Document Type


Degree Name

Master of Science in Computer Science


Computer Science and Engineering

First Advisor

Mohan Kumar


New network applications are being created everyday to accommodate diverse user needs. Delivering services to the user in a timely manner taking into account network conditions, resources allocated and network load is a challenge. Multiprotocol Label Switching attempts to overcome best-effort service by providing a method for routing traffic around network congestion, resource reservation and quality of service (QoS) capabilities. IntServ and DiffServ are two other QoS models in use today. IntServ provides per-flow guarantee of quality while DiffServ is based on aggregate service classes. Adaptive Network Service (ANS) is a community of adaptive, collaborating agents residing in the network that aims to provide enhanced performance, better quality of service and improved efficiency of network resources. ANS agents are distributed across the network and are strategically located to provide services and monitor network conditions. ANS agents gather network information in these locations and exchange this information with other agents in the community. Awareness of current network status enables the agent community to efficiently allocate resources and provide services in response to incoming user requests. Sharing information allows agents in one part of the network to be aware of conditions in other parts of the network. ANS uses this information to route user data flows away from congested network areas. The agents also share resource utilization information in different ANS nodes. Routing user connections to different service points based on current resource utilization leads to efficient use of resources. In our implementation we provide a TCP based service for transferring bulk data from a source to a destination. When a user contacts ANS, ANS first determines the best available node to service the user request. The ANS node to service the request is determined before the start of data flow and is selected on the basis of i) least congestion around the ANS node and ii) maximum availability of resources in the ANS node. This thesis describes an architecture for ANS and derives a mathematical model for ANS's bulk data transfer service while presenting simulation results for a various experiments demonstrating the benefits of using ANS. Simulation results and model estimates show that ANS is able to achieve superior data transfer throughput compared to connections that do not use ANS scheme. Simulation results show that use of ANS improves data transfer performance by a factor of 2 in low traffic conditions and by a factor of up to 3.8 times in high traffic conditions. Future work in this direction includes introducing new services into the ANS framework and improving ANS agent intelligence to deliver these services in a user friendly way. We intend to enhance ANS for applications in pervasive computing environments.


Computer Sciences | Physical Sciences and Mathematics


Degree granted by The University of Texas at Arlington