A number of active queue management algorithms have been studied since Random Early Detection (RE... more A number of active queue management algorithms have been studied since Random Early Detection (RED) was first introduced in 1993. While analytical and experimental studies have debated whether dropping/marking should be based on average or instantaneous queue length or, alternatively, based on input and output rates (or queue length slope), the merits and drawbacks of the proposed algorithms, and the effect of load-based versus queue-based control have not been adequately examined. In particular, only RED has been tested in realistic configurations and in terms of user metrics, such as response times and average delays. In this paper, we examine active queue management (AQM) that uses both load and queuing delay to determine its packet drop/mark probabilities. This class of algorithms, which we call load/delay controllers (LDC), has the advantage of controlling the queuing delay as well as accurately anticipating incipient congestion. We compare LDC to a number of well-known active queue management algorithms including RED, BLUE, FRED, SRED, and REM in configurations with multiple bottlenecks, round trip times and bursty Web traffic. We evaluate each algorithm in terms of Web response time, delay, packet loss, and throughput, in addition to examining algorithm complexity and ease of configuration. Our results demonstrate that load information, along with queue length, can aid in making more accurate packet drop/mark decisions that reduce the Web response time.
We investigate the effect of TCP explicit congestion notification (ECN) with a new response strat... more We investigate the effect of TCP explicit congestion notification (ECN) with a new response strategy that is more aggressive in the short term, but preserves TCP long term behavior, without modifying the router marking rate. A less aggressive ECN decrease gives more incentives for end systems to become ECN-compliant, as ECN serves as an early warning sign in this case. Our analysis and simulation results demonstrate the effectiveness of the new algorithm in improving throughput and reducing fluctuations. We model a multiple bottleneck scenario with various types of traffic, and evaluate the effect of a number of parameters, including TCP flavor, increase/decrease parameters, buffer size, and random early detection (RED) parameters.
ABSTRACT As widely adopted technologies tend to be ossified, Internet Service Providers (ISPs) ar... more ABSTRACT As widely adopted technologies tend to be ossified, Internet Service Providers (ISPs) are reluctant to adapt their core infrastructure to accommodate novel technologies. Overlay networks have emerged as a new strategy to suggest ways of introducing disruptive technology under these circumstances. Overlays can be used as a vehicle to implement new functionalities between the core IP networks and end-user computers, without the support from the ISPs. Several types of overlay networks for resilient routing, multicasting, quality of service, content distribution, storage, and object location have recently been proposed. Overlay networks offer several attractive features, including ease of deployment, flexibility and adaptivity, and an infrastructure for collaboration among hosts. Despite this potential, designing efficient, scalable, and secure overlay networks is a challenging problem. Since overlays utilize only unicasts among pairs of end-systems, they are prone to sacrifice performance compared with router-level solutions. We investigate a heuristic overlay multicast approach, which we refer to as Topology Aware Grouping (TAG). TAG exploits underlying network topology data to construct overlay multicast networks. This topology data is overlap in routes from the source among group members. We implement and experiment with TAG on the PlanetLab wide-area platform. Results demonstrate the effectiveness of our heuristic in reducing delays and duplicate packets. We also characterize the performance penalty for overlay multicast trees via experimental data, simulations, and theoretical models. Three overlay multicast protocols are compared with respect to various performance metrics. Overlay multicast routing strategies and host distribution, together with Internet topology characteristics, are identified as causes of the observed phenomena. Our results also quantify the overlay cost at the network-layer. In addition, we explore the potential of cooperation among co-existing, possibly heterogeneous, overlay networks. We investigate a spectrum of cooperative forwarding and information sharing services, and discuss the associated scalability, security, and heterogeneity problems. Motivated by these services, we design Synergy , an overlay internetworking architecture that fosters overlay cooperation. Our architecture promotes fair peer relationships among overlays to achieve synergism. Results indicate that Synergy can improve performance for overlay nodes, while maintaining the autonomy of individual overlay networks.
A number of active queue management algorithms have been studied since Random Early Detection (RE... more A number of active queue management algorithms have been studied since Random Early Detection (RED) was first introduced in 1993. While analytical and experimental studies have debated whether dropping/marking should be based on average or instantaneous queue length or, alternatively, based on input and output rates (or queue length slope), the merits and drawbacks of the proposed algorithms, and the effect of load-based versus queue-based control have not been adequately examined. In particular, only RED has been tested in realistic configurations and in terms of user metrics, such as response times and average delays. In this paper, we examine active queue management (AQM) that uses both load and queuing delay to determine its packet drop/mark probabilities. This class of algorithms, which we call load/delay controllers (LDC), has the advantage of controlling the queuing delay as well as accurately anticipating incipient congestion. We compare LDC to a number of well-known active queue management algorithms including RED, BLUE, FRED, SRED, and REM in configurations with multiple bottlenecks, round trip times and bursty Web traffic. We evaluate each algorithm in terms of Web response time, delay, packet loss, and throughput, in addition to examining algorithm complexity and ease of configuration. Our results demonstrate that load information, along with queue length, can aid in making more accurate packet drop/mark decisions that reduce the Web response time.
We investigate the effect of TCP explicit congestion notification (ECN) with a new response strat... more We investigate the effect of TCP explicit congestion notification (ECN) with a new response strategy that is more aggressive in the short term, but preserves TCP long term behavior, without modifying the router marking rate. A less aggressive ECN decrease gives more incentives for end systems to become ECN-compliant, as ECN serves as an early warning sign in this case. Our analysis and simulation results demonstrate the effectiveness of the new algorithm in improving throughput and reducing fluctuations. We model a multiple bottleneck scenario with various types of traffic, and evaluate the effect of a number of parameters, including TCP flavor, increase/decrease parameters, buffer size, and random early detection (RED) parameters.
ABSTRACT As widely adopted technologies tend to be ossified, Internet Service Providers (ISPs) ar... more ABSTRACT As widely adopted technologies tend to be ossified, Internet Service Providers (ISPs) are reluctant to adapt their core infrastructure to accommodate novel technologies. Overlay networks have emerged as a new strategy to suggest ways of introducing disruptive technology under these circumstances. Overlays can be used as a vehicle to implement new functionalities between the core IP networks and end-user computers, without the support from the ISPs. Several types of overlay networks for resilient routing, multicasting, quality of service, content distribution, storage, and object location have recently been proposed. Overlay networks offer several attractive features, including ease of deployment, flexibility and adaptivity, and an infrastructure for collaboration among hosts. Despite this potential, designing efficient, scalable, and secure overlay networks is a challenging problem. Since overlays utilize only unicasts among pairs of end-systems, they are prone to sacrifice performance compared with router-level solutions. We investigate a heuristic overlay multicast approach, which we refer to as Topology Aware Grouping (TAG). TAG exploits underlying network topology data to construct overlay multicast networks. This topology data is overlap in routes from the source among group members. We implement and experiment with TAG on the PlanetLab wide-area platform. Results demonstrate the effectiveness of our heuristic in reducing delays and duplicate packets. We also characterize the performance penalty for overlay multicast trees via experimental data, simulations, and theoretical models. Three overlay multicast protocols are compared with respect to various performance metrics. Overlay multicast routing strategies and host distribution, together with Internet topology characteristics, are identified as causes of the observed phenomena. Our results also quantify the overlay cost at the network-layer. In addition, we explore the potential of cooperation among co-existing, possibly heterogeneous, overlay networks. We investigate a spectrum of cooperative forwarding and information sharing services, and discuss the associated scalability, security, and heterogeneity problems. Motivated by these services, we design Synergy , an overlay internetworking architecture that fosters overlay cooperation. Our architecture promotes fair peer relationships among overlays to achieve synergism. Results indicate that Synergy can improve performance for overlay nodes, while maintaining the autonomy of individual overlay networks.
Uploads