ECOOP'2004 CADS Workshop

ECOOP'2004 Workshop on

WS 7 : Communication Abstractions for
Distributed Systems
Oslo, Norway June 14, 2004

Antoine Beugnard(Chair): antoine.beugnard@enst-bretagne.fr	ENST-Bretagne, Brest, France
Eric Jul (Co-chair) : eric@diku.dk	DIKU, University of Copenhagen, Copenhagen, Denmark
Laurence Duchien duchien@lifl.fr	Université de Lille, France
Ludger Fiege fiege@gkec.tu-darmstadt.de	Darmstadt University of Technology, Germany
Robert Filman rfilman@arc.nasa.gov	NASA Ames Research Center, USA
Salah Sadou salah.sadou@iu-vannes.fr	Valoria, Université de Bretagne Sud, France

Abstract

Applications have become increasingly distributed. Distribution complicates systems building and exacerbates problems such as dealing with failure, and providing security, quality of service, reliability, and manageability.

System development is eased by abstraction and modeling. How to model distributed systems? Distributed systems can be understood as communicating objects. To tackle the problems of building distributed systems, it is useful to focus on the abstract issues of inter-component communication. Examples of distributed communication mechanisms include messaging systems, remote procedure calls, distributed objects, peer-to-peer and publish-and-subscribe. Within any such paradigm, there are many opportunities for specialized and detailed engineering decisions. While mechanisms such as these are a good foundation for dealing with the problems of distribution, there remain many issues about how to mold these ideas to deal with the problems of real systems.

At the previous ECOOP workshops, we identified some problems (security, privacy, partial failure, guaranteeing quality of service, run-time evolution, meta-object protocols, and ordering of events) that are important concerns of any communication abstraction. The goal of this workshop is to contrast and compare communication abstractions for distributed systems. Participants will be asked to submit a position paper on some aspect of communication abstractions for distributed systems. To focus the group’s discussion, this year we consider the distributed aerospace information problem, described in the call-for-papers. Prospective participants are requested to relate their contribution to some facet of that that problem. The workshop itself will consist of short presentations, discussion of those presentations, and division into smaller topic study groups.

Call for participation

We are interested in papers reporting practical experiences relating both benefits and obstacles in using communication abstractions. The word abstraction should be understood as “higher level.” (Communications should also be understood as communications among machines, not human-machine interface.) The main questions are what are the possible abstractions, what are their properties, how to implement them. At previous ECOOP workshops (Communication Abstractions for Distributed Systems (2003) and The Next 701 Distributed Object Systems (2002)) we studied some problems inherent to distribution, such as security, partial failure, guaranteeing quality of service, run-time evolution, and considered what tools an object system might supply to help address them. Technologies included grouping objects into components, immutable objects, application-level protocols, reflection (both introspection and reification), and event-ordering.

This year, to help increase the coherence of the discussion, we prefer position papers that speak to the distributed information issues of modernizing Airspace Systems (http://www.nas-architecture.faa.gov/Tutorials/NAS101.cfm). In 25 years, we'd like to have every aircraft, counter, terminal, baggage carousel, control tower and gate networked so that information generated by any of these is conveyed to other interested parties. That is, if a plane flies over the Alps and experiences turbulence, then that turbulence information should be communicated to other pilots on the same path. If the turbulence has slowed down the plane's arrival, then the gates and baggage carousels, connecting flights, automobile rentals of the passengers, etc. all bear notification and perhaps rescheduling. Someone studying patterns of Alpian turbulence should find the data in her database, though not with the same alacrity as a pilot flying from Paris to Milan.

Important ilities that proposed organizations need to deal with are efficiency (you can't tell everyone everything), maintainability (you can't turn off the air system), evolvability (you don't know all the future applications of the data), scalability (this is a big system), reliability (for obvious reasons), quality of service (getting important information to its destinations quickly and deferring the unimportant) and security (all the issues of keeping fake messages out of the system, and also federated security: airlines may be willing to share some information with the governing authorities and the respective airplane manufacturers, but not with each other).

More generically, possible topics for communication abstractions include:

Communication abstractions themselves, such as synchronous and asynchronous messages, publish-and-subscribe, peer-to-peer, and group and broadcast communications, and parts of communication abstractions, such as coordination, mobility, migration, persistence, security, privacy and reliability techniques.
Embodiments of communication abstractions, such as middleware services, mediation, glueware, communications-centric programming languages, communication frameworks, and communication components, such as run-time system and protocol evolution.
Dealing with failure in communications and communicants, including transactions and recovery mechanisms.
Semantic issues of communications, such as ordering of events.
Conceptual organizations for communications, such as design patterns for communication and distribution, and composition mechanisms for protocols.

Our long-term goal is to define and refine abstractions that address some of these problems and other like them. What are the right abstractions, APIs, development methods, reasoning systems, and tools for building the next generation of Distributed Object Systems?

This workshop aims to foster discussion during the workshop. The workshop is not a mini-conference. Position papers, not to exceed 6 pages in length, are solicited by April 5, 2004. Papers based on experience with the above issues are particularly welcome.

Please send positions papers electronically in PDF or Postscript format to Eric Jul at eric@diku.dk and Antoine.Beugnard@enst-bretagne.fr by April 5, 2004. Notification of acceptance will be given by April 26.

A maximum of 20 participants will be selected on the basis of the submitted material. Submitted position papers should include a 100 word abstract and a set of relevant keywords. The number of participants per position paper is limited to 2.

Springer-Verlag will publish the ECOOP 2004 Workshop Reader as an Lecture Notes in Computer Science LNCS volume. This book will include a report for each workshop. The organizers will write the report, in collaboration with the participants of the workshop. The organizers will produce a report that provides a summary of the workshop with the major issues discussed and the conclusions of the working groups (if applicable). The report will also include the current research being carried out in the area and open research directions on the workshop themes.