Panel - “Middleware for Real-Time Distributed Objects: Needs and Requirements from Different Application Domains” Panelists: Bran Selic - Rational - Canada - bselic@rational.com Richard Turner - DoD - USA - Rich.Turner@osd.mil Rob van den Berg - Philips - Netherlands - rob.van.den.berg@philips.com Carlo Batini - AIPA - Italy - batini@Aipa.it Greg Bolella - SUN - USA - Luiz Bacellar - United Technologies Research Center - USA - bacelllf@utrc.utc.com Sanjay Razdan - Tridium - USA - srazdan@mail.tridium.com Panel Organizer: Carlos E. Pereira - United Technologies Research Center - USA - pereirce@utrc.utc.com (on leave from Federal University of Rio Grande do Sul - Brazil) Motivation: The main goal of this panel was to evaluate the needs and requirements that different application domains impose on middleware based on distributed real-time objetcs. Panelists were specialists from different application domains, with a large experience on the development of complex real-time distributed systems. Richard Turner (software intensive military applications) The US Department of Defense plans a large number of network-centric systems to implement warfighting, C4ISR, and support functions. These systems will need well-behaved, predictable, robust, and high-speed middleware to meet their mission critical requirements. While there is similarity in the requirements of military and commercial systems (particularly in the transportation and medical domains), military systems are generally longer-lived, more complex, require more varied interfaces, and must operate in a more hostile environment than commercial systems. There is currently some concern about the reliability and performance of existing real-time middleware. Research in this area is a priority, particularly where middleware can support rapid reconfiguration, multi-level security, and support for widely varying legacy components. Bran Selic (telecommunication) Most telecommunications software can be grouped into one of two broad categories. In one case we have software that is involved in actual information transmission. This includes relatively simple functions such as setting up and tearing down of virtual circuits, routing of packets through protocol stacks, etc. Transmission typically requires low latency and high bandwidth implying that software overheads must be minimal. The second category, often referred to as Operations, Administration, and Maintenance (OA&M), is used to control the actual telecommunications network and equipment. This software is usually much more complex, but generally requires less bandwidth and is much more delay tolerant than transmission software. COTS middleware—mostly CORBA based—is being applied with significant success in telecom systems but almost exclusively for OA&M. This is primarily due to the following reasons: • Throughput and delay are still inadequate for most transmission functions (although the situation is improving steadily) • Most COTS middleware assumes a heavyweight high-overhead threading model with unacceptable context switching times (most telecom systems implement very lightweight concurrency) • The majority of telecom systems are based on proprietary hardware and software. This requires COTS middleware to be ported, which may be costly and difficult (thereby eliminating the primary benefits of using COTS software). • Telecom systems in general have extreme availability and reliability requirements – something that is not supported by most COTS middleware • It is necessary for middleware to respect and, if necessary, enforce end-to-end delay constraints; however, this requires mechanisms that are not found in most COTS middleware Clearly, these shortcomings must be remedied before COTS middleware will be used in transmission software. A particularly useful feature for this purpose would be to provide reflective interfaces on middleware. The Proceedings of Sixth International Workshop on Object-Oriented Real-Time Dependable Systems (WORDS01) 1530-1443/01 $17.00 ' 2001 IEEE