Research Plan for My PhD Program
Postboks 70, 2027 KJELLER
Tlf: 64 84 47 00
Fax: 63 81 81 46
In order to develop distributed multimedia applications in heterogeneous end-systems and network environments, flexible and adaptive frameworks are necessary. End-systems range from handheld devices such as Personal Digital Assistants (PDAs) and cellular phones to personal computers and servers. The processing capacity differs substantially. The networks connecting these devices and machines range from GSM data-service, Ethernet LAN to ATM. The QoS delivered differs for characteristics as for instance throughput, delay, and error rate.
Multimedia applications often use group communication, examples include video conferencing and distance education. Connection management and signalling protocols for group communication must scale from small working groups with few (e.g. 2-10) members, up to conferencing applications with thousand receivers.
Today’s Internet is built on the DARPA protocol suite (TCP/IP), with IPv4 and IPv6 as the enabling infrastructure for higher-level protocols as TCP, UDP, RTP and so on. The existing Internet fails to provide QoS guarantees, therefore work with the Integrated Services (IntServ) and Differentiated Services (DiffServ) is currently undertaken in the IETF. Efforts for IntServ and DiffServ to co-exist in next-generation Internet have been initiated [IERD][IENU].
Current QoS research activities are focusing on various aspects that are relevant during the lifetime of an association between communicating peers. However, the OSI Reference Model includes also QoS parameters at the transport layer that are relevant for connection establishment (i.e., connection establishment delay, connection establishment failure probability) and connection release (i.e., connection release delay, connection release failure probability). Different mechanisms for connection establishment (and release) like no connection set-up, implicit connection set-up, two-way handshake, and three-way handshake lead to different QoS. For example, low latency requirements cannot appropriately supported by a connection set-up with three-way handshake. The same arguments are valid for connection release and for re-configuration during the lifetime of a connection. Connection re-configuration is necessary to adapt communication protocols to changes in network (e.g., congestion or resolution of a former congestion) and end-systems (e.g., end-system load changes drastically) and can roughly be seen as a combination of connection release and establishment. Re-configuration might be performed with loss of protocol data units in a continuous stream or without it. It is important to enable the application to determine which mechanisms are applied for connection set-up, re-configuration and release with appropriate QoS parameters.
The starting point of this project are the experiences made with connection management in Da CaPo [DCP] and in the Multicast Communication Framework (MCF) [MFC]. It has to be investigated how far traditional protocol mechanisms (e.g., those in TCP) can be de-coupled from connection management. Further steps include the development of a framework for flexible connection management, specification and implementation of a connection management or signalling protocol, and the implementation of a demonstrator for the framework.
A flexible protocol system allows the dynamic selection, configuration and reconfiguration of protocol modules to dynamically shape the functionality of a protocol to satisfy specific application requirements and/or adapt to changing service properties of the underlying network. In the following we refer to such a system as a flexible end-to-end protocol. The basic idea of flexible end-to-end protocols is that they are configured to include only the necessary functionality required to satisfy the application for the particular connection. This might even include filter modules to resolve incompatibilities among stream flow endpoints and/or to scale stream flows due to different network technologies in intermediate networks. The goal of a particular configuration of protocol modules is to support the requested QoS for requested connections. This will include point-to-point, point-to-multipoint, and multipoint-to-multipoint connections.
There are several public domain packages for flexible protocol systems available, including HORUS/Ensemble and Da CaPo. Da CaPo is used in the MULTE-ORB, and is currently under re-design and implementation.
The existing CORBA standard uses an “black box” approach, in the sense that only external interfaces are specified. The internal interfaces between components are not specified, and different CORBA implementations have different designs. This is based on historic development, where ORB vendors transformed existing products to conform with CORBA. For the next generation middleware the “white box”, or open implementation approach is natural. The new paradigm called component-oriented development may also be used to design the middleware itself.
The CORBAng project advocates the use of a reflective architecture for next generation middleware [CNG98] [CNG99]. A binding framework has been specified and the work with the communication framework is in progress. The signalling framework in this project plan will be aligned with the work in CORBAng.
Work with architectural issues and development of a framework for flexible connection management is combined with evaluation of existing prototypes and extensions to these prototypes. At the end a flexible signalling framework will be implemented, as an instantiation of the reflective architecture and the flexible signalling framework.
The signalling protocol will be specified using PROMELA and validated using the SPIN tool [HOLZ], before the implementation starts. Expressing the protocol in a formal language guides the design process, and the validation will show the correctness of the protocol before implementation, “real-life” testing and measurements is initiated.
In order to design and implement a flexible signalling framework, we need a middleware platform supporting specification of QoS and mapping to resources and functionality in the communication system.
The MULTE-ORB [ICDCS] incorporates Da CaPo [TK99] and has been enabled to let applications specify QoS requirements [RTB99] and communicate these QoS parameters to Da CaPo. Da CaPo uses these requirements to tailor a suitable protocol stack for the communication.
Further work will include:
A prototype with these characteristics will support the research and exploration of the flexible connection management.
To navigate in the research field a survey of open, flexible signalling will be performed. Examples of surveys to build on is Stiller [BS95] and Doeringer et.al. [WD94].
The connection management and signalling framework need to be integrated in the reflective architecture, and will be designed as an instantiation of this architecture. Coordination with the binding framework and the communication framework needs special attention.
After the survey, analysis and the initial specification of the signalling framework, the first version of the thesis proposal will be written.
When it comes to evaluation, experiments and testing of existing connection management protocols and approaches, RSVP and ATM signalling are candidates for investigation. Possible changes required will be identified in field trials and evaluation.
After the thesis proposal has been written, and experiments with existing connection management approaches has been conducted. The design and implementation of the signalling framework will start. A design of the signalling protocol will be specified in a formal language and tested using a validator tool. The demonstrator application will be aligned with the OMODIS-LoD application prototype and the MULTE demonstrator at FFI.
Tasks and goals under short term goals will be reached early spring 2000. The specification and design of the flexible signalling framework, together with survey and writing of the initial version of the Thesis Proposal will start in spring 2000 and end early spring 2001. The Thesis Proposal have deadline in the middle of the autumn semester 2000. Please refer to Figure 1 for a timeline and graphic visualization.
Figure 1: High-level time schedule
After finishing the design, implementation of parts of the framework will start. Experiments and evaluation of the framework will be done in parallel, giving feedback to the implementation and design. Thesis writing and final evaluation of the work take place in spring semester 2003.
[AVS] Object Management Group: “Control and Management of Audio/Video Streams”, in “CORBAtelecoms: Telecommunications Domain Specifications”, version 1.0, OMG Document formal/98-07-12, June 1998
[CNG98] G.S. Blair, G. Coulson, P. Robin, and M. Papathomas: “An Architecture for Next Generation Middleware”, Internal Report MPG-98-27, Department of Computing, Lancaster University, 1998
[CNG99] Frank Eliassen, Gordon Blair, Fabio Costa, Katia B. Saikoski, Geoff Coulson, Vera Goebel, Øivind Hansen, Tom Kristensen, Thomas Plagemann, and Hans Ole Rafaelsen: “Next Generation Middleware: Requirements, Architecture, and Prototypes”, to be submitted: Future Trends of Distributed Computing Systems (FTDCS ’99), Cape Town, South-Africa, December 1999
[DCP] Thomas Plagemann: “A Framework for Dynamic Protocol Configuration”, PhD Thesis, Swiss Federal Institute of Technology (Diss. ETH No. 10830), Zurich, Switzerland, September 1994
[ICDCS] Tom Kristensen and Thomas Plagemann: “Enabling the Object Request Broker COOL for Flexible QoS Support”, submitted to The 20th International Conference on Distributed Computing Systems (ICDCS2000), Taipei, Taiwan, April, 2000
[IERD] Y. Bernet, R. Yavatkar, P. Ford, F. Baker, L. Zhang, M. Speer, R. Braden, and B. Davie: “Integrated Services Operation Over Diffserv Networks”, IETF Internet Draft (draft-ietf-issll-diffserv-rsvp-02.txt), June 1999
[HOLZ] Gerard J. Holzmann: “Design and Validation of Computer Protocols”, Prentice Hall Software Series, ISBN 0-13-539925-4, Prentice Hall, 1991
[MFC] Daniel Bauer: “A Multipoint Communication Architecture for End-to-End Quality of Service Guarantees”, PhD Thesis, Swiss Federal Institute for Technology (Diss. ETH No. 12163), Zurich, Switzerland, April 1997
[TK99] Tom Kristensen: “Extending the Object Request Broker COOL with Flexible QoS Support” (in Norwegian), Master Thesis, Center for Technology at Kjeller (UniK), University of Oslo, August 1999
[BS95] Burkhard Stiller: “A Survey for UNI Signaling Systems and Protocols for ATM Networks”, Computer Communication Review, ACM SIGCOMM, Vol 25, No 2, April 1995
[RTB99] Ragnvald T. Blindheim: ““Extending the Object Request Broker COOL with Flexible QoS Support” (in Norwegian), Master Thesis, Center for Technology at Kjeller (UniK), University of Oslo, February 1999
[WD94] W.A. Doeringer, H.D. Dykeman, M. Kaisenwerth, B.W. Meister, H. Rudin, and R. Williamson: “A Survey of Light-Weight Protocols for High-Speed Networks”, chapter 1 in: “High performance networks. Technology and protocols”, Kluwer Academic Publishers, 1994