Note: Un des partenaires de cette proposition de collaboration n'étant pas francophone, la description de ce projet a été rédigée en Anglais.

Keywords: Internet, Mobility, QoS, IntServ/DiffServ, Mobile IP, TCP in Mobile Environment, Active Network.

Collaboration Proposal

The main drawback of the current Internet is the lack of Quality-of-Service (QoS) support. QoS support, however, is essential for business and electronic commerce applications as well as for real-time applications such as Internet Telephony (also known as IP telephony) and on-line video retrieval. During the last years the Internet community spent many efforts to develop an Internet QoS architecture based on the Integrated Services (IntServ) model and based on the Resource Reservation Setup Protocol (RSVP). However, the IETF RSVP working group stated that RSVP and the integrated services approach can not be deployed in large-scale Internet backbones due to scaling problems.

Differentiated Services (DiffServ) are a new approach for Quality-of-Service support in the Internet. Differentiated services are expected to provide a better QoS than existing services today but simultaneously keeping the implementation complexity for ISPs reasonably low. However, Differentiated Services has in its current form the following problems, which are significant in mobile environments.

1. QoS can only be provided if the backbone networks of the Internet Service Providers are well designed and provisioned. However, network provisioning is a relatively complex task. Network provisioning especially becomes very difficult in highly dynamic environments, in particular in networks where the location and the QoS requirements of the end systems may change very quickly. In particular, this is true for mobile multimedia end systems. A challenging task is therefore network provisioning in order to support differentiated services for mobile systems connected to the Internet using mobile IP.
   
   
2. If a user wants to change his service level agreement, the changes must be communicated to the ISP using conventional communication mechanisms such as fax, phone or WWW forms. After indicating the changes, the network nodes have to be reconfigured manually by a network operator within one or more ISP domains. This results in significant delays in the range of minutes, hours or even days. Therefore, DiffServ based on static service level agreements is not suited for dynamically changing communication requirements in the Internet. Highly dynamic environments and the static bandwidth allocation concept of DiffServ is contradictory. The goal, however, should be to support dynamic service level renegotiations in order to allow the reaction of the dynamics of mobile users. 
   
   
3. Another problem with combining DiffServ and Mobile IP is that the destination home address might not be visible in the IP packet header´s destination address making difficult to identify flows of Mobile users. This is true for both IPv6 (the destination home address might be moved to the IPv6 routing header in the first part of the transmission path) and IPv4 (the original IPv4 packet is encapsulated for transmission over a mobile IP tunnel).

The last problem must be solved by enhancing the functionality of DiffServ routers in order to support flows of mobile IP systems. This requires the modification of flow classification procedures within DiffServ boundary routers.

The first problem can be solved by careful network planning based on statistics and predictions about the bandwidth requirements and the movements of mobile end systems. Another solution would be the deployment of a signaling mechanism within the mobile end systems. In this case, the bandwidth requirements in a DiffServ network can be derived more accurately using the signaling information. However, this also requires the possibility of dynamic router reconfigurations. The approach of a signaling mechanism for Diffserv also solves the second problem mentioned since it allows dynamic reconfigurations of the DiffServ network according to the current needs. Several options may be used as a basis for a DiffServ signaling approach:

1. RSVP can be used as a signaling protocol within the end systems. Edge routers of the DiffServ network analyze these RSVP messages and adapt the DiffServ configuration parameters dependent on the collected RSVP information. However, an out-of-band signaling approach such as RSVP might create some problems for bursty and short-lived connections. A great advantage of this approach, however, is the upcoming support of RSVP in many operating systems.  
   
   
2. Another approach is the definition of special IPv6 options / extension headers in addition to the ones under development for mobile IPv6. This has the advantage of integrating the signaling information and the IPv6 data packets and avoiding the out-of-band approach. 
   
   
3. From the latter option, the approach of active networking can be derived. In the case of active networking not only passive option data are included in the IPv6 packets but complete programs or scripts. These active elements might be used to reconfigure the routers which are passed by the IPv6 data packet. In particular, DiffServ parameters can be reconfigured dynamically by these active elements.

One goal of the collaboration is the conception and the evaluation of an appropriate mechanism to support the of combination of the Diffserv approach and mobile IP. The evaluation of our proposal will be performed using the Berkeley NS (Network Simulator) simulator.

Within this collaboration, the partners want to benefit from each other. While INRIA has much experience in the area of mobile networking (an particular in mobile IP), University of Berne and ENST are running several research projects covering several DiffServ topics, which are funded by public funding organizations and industrial companies. Therefore, the research interests are complementary with a very high degree.

A second goal of this project is to study the impact of this DiffServ model on the existing transport protocols such as the TCP protocol in a mobile environment. We will study protocols such as I-TCP or Snoop-TCP in a DiffServ network. We will propose some improvements that can be brought by the known characteristics of the wireless and mobile environment and its different layers. Our proposals will be evaluated through simulations.