Section: Scientific Foundations
Quality of Service and Transport layer for Future Networks
Participants : Pascale Vicat-Blanc Primet, Laurent Lefèvre, Sébastien Soudan, Romaric Guillier, DInil Mon Divakaran, Guilherme Koslovski, Isabelle Guerin-Lassous, Rémi Vannier.
Congestion control is the most important and complex part of a transport protocol in a packet switched shared network. The congestion control algorithm is then a key component which has to be considered to alleviate the performance problems in the future networks environments. TCP has shown a great scalability in number of users, but not in link capacity and link diversity. For example, TCP performance can be very low and unstable in data-center applications and interactive communications within high speed long distance networks infrastructures, like lambda grids environments. The conservative behavior of TCP with respect to congestion in IP networks is at the heart of the current performance issues faced when the traffic load is highly dynamic. On the application side, one can observe that traditional applications were originally characterized by very basic communication requirements related to performance, reliability and order. The rapid deployment of new heterogeneous network technologies has pushed the development of an important number of new multimedia applications presenting complex requirements in terms of delay, bandwidth constraints and tolerance to losses. These applications need specific mechanisms to adapt to network congestion or changing medium conditions. To solve this problem, protocol enhancements and alternative congestion control mechanisms have been proposed for very high speed optical networks, wireless networks and for multimedia applications (see PFLDNET conference series). Most of them are now implemented in current operating systems, but these protocols are not equivalent, and not all of them are suitable for all environments and all applications, moreover they may not cohabit well. Since a couple of years, the evaluation and comparison of new transport protocols received an increasing amount of interest (see IRTF TMRG and ICCRG groups). However, TCP and other alternatives are complex protocols with many user-configurable parameters, and a range of different implementations. Several aspects can be studied, and various testing methods exist. The research community recognizes that it is important to deploy measurement methods so that the transport services and protocols can evolve guided by scientific principles. Researchers and developers need agreed-upon metrics, a common language for communicating results, so that alternative implementations can be compared quantitatively. Users of these variants need performance parameters that describe protocol capabilities so that they can develop and tune their applications. Protocol designers need examples of how users will exercise their service to improve the design.
As the Internet has evolved from a research project into a popular consumer technology, it may not be reasonable to assume that all end hosts would fairly cooperate. Indeed, concerns were raised that the recently started deployment of non-IETF-approved high-speed TCP variants could lead to an "arms race" that would eventually have a detrimental effect on the overall performance of the Internet. As another example, commercial Internet accelerators can provide better performance for a single user at the expense of other users. In the future, expecting billions of Internet devices to fairly cooperate to prevent network congestion is overly optimistic. New bandwidth sharing approach have to be investigated.
Flow scheduling [3] , based on the in-advance knowledge of resource requirement of an application or online estimation of these requirements can be applied. Signaling or real time flow analysis and also scalability issues have to be explored. Distributed and lagrangian relaxation-based solution for bandwidth sharing is also an interesting approach in Future Internet. This approach addresses well the dynamic feature, due to node mobility or traffic variation. However, some problems remain open. First, the sharing models are often very complex to compute, while still being inaccurate. Second, some parameters of allocation algorithms based on lagrangian relaxation are difficult to set, and are often obtained by trial ad error; and hence not optimized. Finally, the proposed solutions are often tested on home-made simulators that are far from being realistic.
We believe some network resource control has to be associated with the end to end flow control approach to offer better quality of experience in Future networks. Network resource control is classified into three time-scales: data, control, and management. Each time-scale corresponds with a level of aggregation : `data' deals with packets; `control' deals with aggregates of packets, i.e. flows; and `management' deals with aggregates of flows. All three time-scales must be addressed, since they all affect the service perceived by users, and the ease and efficiency with which the network can be operated. The current Internet protocols do not well address the control time scale, and do not consider packet aggregates. TCP deals with resource control at data time-scales; while routing protocols, such as BGP and OSPF operate at time-scales of the order of minutes or hours (management time-scale).
We propose to explore packet aggregates and address control timescale in the context of Future Internet not only for performance, but also for manageability and security purposes.
On an other hand, the optical fiber communication will be the predominant mechanism for data transmission in core network and may be also at the access. To address the anticipated terabit demands, dynamically reconfigurable optical networks are envisioned. This vision will be realized with the deployment of configurable optical components, which are now becoming economically viable. To meet the terabit challenge, network designers will enhance core functionality by migrating to devices equipped with tunable transceivers, optical cross-connects and optical add/drop multiplexers. Optical Cross-Connects (OXCs) becomes more and more, cheap, simple and controllable. The control-plane, traditionally in the hand of telco migrates progressively to the customers. Studying the interactions of components required to accomplish the tasks of bandwidth reservation, path computation and network signaling is an other goal.
