This thesis investigates TCP connection management mechanisms in order to understand the behaviour and improve the performance of Internet servers during overload conditions such as flash crowds. We study several alternatives for implementing TCP connection establishment, reviewing approaches taken by existing TCP stacks as well as proposing new mechanisms to improve server throughput and reduce client response times under overload. We implement some of these connection establishment mechanisms in the Linux TCP stack and evaluate their performance in a variety of environments. We also evaluate the cost of supporting half-closed connections at the server and assess the impact of an abortive release of connections by clients on the throughput of an overloaded server. Our evaluation demonstrates that connection establishment mechanisms that eliminate the TCP-level retransmission of connection attempts by clients increase server throughput by up to 40% and reduce client response times by two orders of magnitude. Connection termination mechanisms that preclude support for half-closed connections additionally improve server throughput by up to 18%.

This thesis investigates TCP connection management mechanisms in order to understand the behaviour and improve the performance of Internet servers during overload conditions such as flash crowds. We study several alternatives for implementing TCP connection establishment, reviewing approaches taken by existing TCP stacks as well as proposing new mechanisms to improve server throughput and reduce client response times under overload. We implement some of these connection establishment mechanisms in the Linux TCP stack and evaluate their performance in a variety of environments. We also evaluate the cost of supporting half-closed connections at the server and assess the impact of an abortive release of connections by clients on the throughput of an overloaded server. Our evaluation demonstrates that connection establishment mechanisms that eliminate the TCP-level retransmission of connection attempts by clients increase server throughput by up to 40% and reduce client response times by two orders of magnitude. Connection termination mechanisms that preclude support for half-closed connections additionally improve server throughput by up to 18%.

En aquesta tesi proposem dos esquemes de xarxa amb control d'admissió per al trànsit elàstic TCP amb mecanismes senzills. Ambdós esquemes són capaços de proporcionar throughputs diferents i aïllament entre fluxos, on un "flux" es defineix com una seqüència de paquets relacionats dins d'una connexió TCP. Quant a l'arquitectura, ambdós fan servir classes de paquets amb diferents prioritats de descart, i un control d'admissió implícit, edge-to-edge i basat en mesures. En el primer esquema, les mesures són per flux, mentre que en el segon, les mesures són per agregat. El primer esquema aconsegueix un bon rendiment fent servir una modificació especial de les fonts TCP, mentre que el segon aconsegueix un bon rendiment amb fonts TCP estàndard. Ambdós esquemes han estat avaluats satisfactòriament a través de simulació en diferents topologies de xarxa i càrregues de trànsit. / In this thesis we propose two network schemes with admission control for TCP elastic traffic using simple mechanisms. Both schemes are able to provide different throughputs and isolation between flows, where a flow is defined as a sequence of related packets within a TCP connection. Regarding to the architecture, both use packet classes with different discarding priorities and an admission control that is implicit, edge-to-edge and based on measurements. In the first scheme, measurements are per-flow, while in the second one, measurements are per-aggregate. The first scheme achieves a good performance using a special modification of TCP sources, while the second scheme achieves a good performance with standard TCP sources. Both schemes have been evauated succesfully through simulation in different network topologies and traffic loads.