Opening Address

This, the opening session of Middleware'98, was briefly preceeded by welcome addresses from both Gordon Blair, General Chair, and Nigel Davies, Programme Co-chair. The format of the conference was briefly described and all of the sponsors formally thanked. Particular reference was made to Sony Corporation for generously stepping in as replacement sponsor for the Thursday evening Steamer trip at such short notice.

Session Chair

Morris Sloman is a member of the Distributed Software Engineering Section, Department of Computing at Imperial College, UK. His research interests include tools and techniques for building distributed systems, distributed systems management, mobile multimedia systems and policies for network and distributed systems management.

A. O'Toole
IONA Technologies

The first technical presentation in this session was a keynote address by Annrai O'Toole of IONA Technologies. IONA, which began trading in 1991, is a major proponent of CORBA. Annrai's talk was designed to present their view on CORBA technology, it's primary uses and future development.

While the original idea behind CORBA was to provide "distributed C++", it has developed as a more generic technology. The two primary uses of CORBA are as a platform for developing new network based applications and, perhaps more significantly, for the integration of existing systems. Annrai was keen to point out that there is no such thing as a legacy system, only real working systems, adding that incompatibility is business. This point was substantiated by comments about the millenium bug and how it is not feasible to just replace 30 years of software investment: instead these existing systems have to be adapted.

In Annrai's opinion, two-tier client/server solutions had been a blip in the evolution of computer technology and argued that three-tier solutions were the most scalable. He added that three-tier computing was born out of mainframe systems, to date one of the most scalable computing environments we have developed. Delegates were informed of how CORBA was currently evolving, not least by borrowing the best ideas from other systems and current research, the principle aim being to achieve maximum flexibility.

The address concluded with comments that middleware was there to tame the network, provide an application server platform and enable service based architectures. Additionally, such needs to be easy to use and, above all, "ensure software works together". Such a solution could not come from Microsoft, Oracle or IBM, but is offered by CORBA.

In response to questions from the floor, Annrai added that his slides would be made available on the internet and that there was plenty of scope for competition in the middleware marketplace.

S. Wheater, S. Shrivastava and F. Ranno
Newcastle University, UK

Presented by Stuart Wheater

Internet applications feature large sequences of tasks with many inter-dependencies, requiring dependable scheduling, dynamic workflow reconfiguration and tolerance in the face of failure. Stuart suggested that to engineer such applications it was necessary to address design requirements of interoperability, scalability, dependability and separating work from flow. To this end, a computational model for dealing with workflows that offered two sets of complementary services with CORBA IDL interfaces was described. The concepts of compound tasks and generic tasks as mechanisms for achieving flexible task composition were discussed. The presentation ended by mentioning that C++ and Java implementations of the architecture existed, then by listing proposed future work.

A couple of questions were raised by the floor. First, could a task be both compound and genesis at the same time? Stuart's answer was that a genesis task could start a compound task, implying the two were distinct. Secondly, it was questioned whether the choice of CORBA had influenced the design of the task model. Stuart replied that it hadn't as such, although this had meant some refinements. He added that object references were passed around in the architecture rather than objects themselves.

A. Schill and C. Mittasch
TU Dresden, Germany

Presented by Christian Mittasch

Christian began by outlining this research, which has been so far been the subject of three years study. He commented that his focus was on business information systems in general rather than workflow management systems. Business Objects were introduced as software components with common attributes of identification, semantic description and reuse and that their level of abstraction facilitated convergence. A framework named BPAFrame was outlined and it was argued that the combination of Business Objects and CORBA within this enabled generic solutions.

Panel

Chair: Morris Sloman
AOT: Annrai O'Toole
SW: Stuart Wheater
CM: Christian Mittasch

Question: There has been a lot of research into workflow, what contributions do the speakers offer?
CM: Persistent state of objects orthogonal to the tasks of distributed computing. Three-tier architectures offer better solutions.
SW: Autonomous workflows.

Question: Composition objects and concurrent tasks are necessary. Can you support these?
SW: Yes.
CM: Our system was tested with up to 10,000 concurrent workflows. Business Objects act autonomously in the case of failure.

Comment: Systems people often forget to analyse systems in advance of deployment.
SW: The simplicity of the computational model determines whether analysis is straightforward. We don't yet offer a full set of tools.
CM: The high level of abstraction between workflows and workflow types helps to address this.
AOT: We see this through the management of running systems all the time. There are no easy solutions and lot's of research still to be done. Workflow is only a subpattern of a larger thing.

Question: Isn't it necessary to manage many different configurations simultaneously?
AOT: I totally agree. Dynamic system component loading and unloading is the subject of our current work. To do this you need to provide an underlying engine which is flexible.
SW: Our advantage is that we keep a repository of workflow types, so we can support this.

Question: Tolerating failure doesn't mean just machine failures. What about incorrect workflow designs or unforseen business model changes?
SW: We provide CORBA interfaces which allow the monitoring, and thus reconfiguration of, components.

Question: We did a recent survey and everyone uses NT, not CORBA. How do you explain this?
AOT: I've no problem with that. If you want just NT, use NT. But NT doesn't work in a heterogeneous environment. CORBA is the cross-platform solution. NT is part of the problem, not the solution. If you don't tell your vendor that you need hetergeneous operation, he'll sell you an off-the-shelf solution that doesn't support it.

Question: What do you think of messaging solutions?
AOT: There's growth in that area. But customers don't see the publish and subscribe mechanisms of IP multicast and so forth. The asynchronous point is only a side issue: solutions are actually bought for their workflow systems.