Multipath TCP (MPTCP) is an extension to regular TCP that exploits the idea of resource pooling by transmitting the data of a single TCP connection simultaneously across multiple Internet paths between two end hosts. Operating system vendors are traditionally in the key position to facilitate the deployment of new functionality, such as MPTCP, to user devices, but their motivation to invest in such enhancements is not self-evident. A scenario in which one party has the capability to deploy software changes on both the mobile devices and on the content servers helps to understand the potential first-mover advantages associated with the deployment of a new IETF standard protocol. In this study, we have built a quantitative techno-economic model to estimate the implementation costs for a content provider selling application downloads who is also able to implement MPTCP on both ends and compare them against the revenue MPTCP generates. The results suggest that even a relatively small increase in the number of downloads could make the business case profitable within five years.
The paper deals with Quality of Service (QoS) in e-commerce Web servers. We consider request service control in a Business-to-Consumer (B2C) Web server system to ensure two business goals: high revenue of the e-retailer and high QoS for key customers with different business values. A problem of optimization of a multi-stage decision process in the context of achieving these goals is formulated. Some simulation results of the system performance under a novel admission control (AC) and scheduling algorithm are discussed.
Re-inserted ECN (re-ECN) is a proposed TCP/IP extension that informs the routers on a path about the estimated level of congestion. The re-ECN protocol extends the Explicit Congestion Notification (ECN) mechanism and reinserts the obtained feedback into the network. This exposure of congestion information is a new economic traffic management mechanism that enables the network to share the available capacity more equally and to police the compliance of congestion control through e. g. a per-user congestion limitation. This paper studies performance implications of the re-ECN mechanism. Our evaluation is based on simulations with an own re-ECN implementation in the Linux TCP/IP stack. Our results also confirm that congestion exposure generally works. But we also showthat traffic characteristics such as the round-trip time (RTT), flow sizes, as well as the selection of congestion control algorithms have a significant impact on the congestion exposure information. These non-trivial effects have to be taken into account when using the re-ECN information as input parameter for congestion control mechanisms in end-systems or for routing/policing inside the network. Comparable results have not been published so far.
The Internet is founded on a very simple premise: sharing! Shared communications links are more efficient than dedicated connections that lie idle much of the time. Hence the rules we use for sharing are extremely vital for the healthy operation of the Internet ecosystem and directly affect the value of the network to its users. It becomes a great paradigm of merging the disciplines of computer science and economics, and presents a great number of challenges to the Internet research community. In this talk we will discuss a number of questions like: what is wrong with today’s Internet sharing technologies? Are these consistent with economics? More specifically, is TCP (Transmission Control Protocol) sensible from an economic point of view? Which network sharing technologies justify end-toend from an economics perspective? What is required to make P2P (peer-topeer) a blessing instead of a curse? Are there bad applications or just inefficient combinations of sharing technologies and pricing schemes? This keynote will survey some of these systems management issues and the emerging approaches to solve them.
The amount of peer-to-peer (P2P) traffic and its inefficient utilization of network resources make it a high priority for traffic management. Due to the distributed nature of P2P overlays, a promising approach for a management scheme is to change the client behavior and its utilization of overlay connections. However, from the viewpoint of a client, there are different categories of overlay connections. In this paper, we discern between these different types of overlay connections in BitTorrent, the currently most popular P2P file-sharing application. A simulation study of BitTorrent and a video-streaming derivate called Tribler provides insights into the usage of these types of connections for data exchange. Thus, traffic management based on client behavior can be optimized by efficiently targeting connections which carry the most traffic. We also show the implications of these results for locality-awareness mechanisms such as Biased Neighbor Selection and Biased Unchoking.
