Christoph Liebig

Event composition in time-dependent distributed systems

Many interesting application systems, ranging from workflow management and CSCW to air traffic control, are event-driven and time-dependent and must interact with heterogeneous components in the real world. Event services are used to glue together distributed components. They assume a virtual global time base to trigger actions and to order events. The notion of a global time that is provided by synchronized local clocks in distributed systems has a fundamental impact on the semantics of event-driven systems, especially the composition of events. The well studied 2g-precedence model, which assumes that the granularity of global time-base g can be derived from a priori known and bounded precision of local clocks may not be suitable for the Internet where the accuracy and external synchronization of local clocks is best effort and cannot be guaranteed because of large transmission delay variations and phases of disconnection. We introduce a mechanism based on NTP synchronized local clocks with global reference time injected by GPS time servers. We agree that timestamps of events can be related to global reference time with bounded accuracy and propose that event timestamps are modeled using accuracy intervals. We present algorithms for event composition and event consumption which make use of accuracy interval based timestamping and illustrate the problems that arise due to inaccuracy and message transmission delays.

DREAM: Distributed Reliable Event-Based Application Management

New applications and the convergence of technologies, ranging from sensor networks to ubiquitous computing and from autonomic systems to event-driven supply chain management, require new middleware platforms that support proactive event notification. We present a system overview and discuss the principles of Dream, a reactive middleware platform that integrates event detection and composition mechanisms in a highly distributed environment; fault-tolerant and scalable event notification that exploits a variety of filter placement strategies; content-based notification to formulate powerful filters and concept-based notification to extend content-based filtering to heterogeneous environments; middleware-mediated transactions that integrate notifications and transactions; and scopes, which are administration primitives for both deployment- and runtime configurability, as well as for the management of policies. We discuss four prototypes that were implemented as proof-of-concept systems and present lessons learned from them.

Middleware mediated transactions

Middleware mediated transactions (MMT) integrate message-oriented transactions and distributed object transactions. MMT are suggested as an evolutionary and integrative approach to support reliable and flexible interactions between heterogeneous and autonomous components, which is a major challenge in enterprise application integration. MMT offer the ability to combine communication of messages and notifications with conventional transactional object requests. Thus MMT introduce the flexibility of mediated interactions with respect to topology, binding, time-dependencies and content transformation into distributed object transactions. MMT are characterized by coupling modes to control if notifications become visible immediately or are dependent on the transaction status, to include mediators as transaction participants, and to distinguish between message delivery and processing, as well as vital and non-vital participants. Furthermore, coupling modes interrelate different distributed transaction contexts of publishers and subscribers. This paper introduces the concept of MMT and presents two system prototypes implementing MMT, the dependency-spheres middleware service and the X/sup 2/ TS middleware service.

Integrating Notifications and Transactions: Concepts and X2TS Prototype

Event-based architectural style promises to support building flexible and extensible component-oriented systems and is particularly well suited to support applications that must monitor information of interest or react to changes in the environment, or process status. Middleware support for event-based systems ranges from peer-to-peer messaging to message queues and publish/subscribe event-services. Common distributed object platforms restrict publishing events on behalf of transactions to message integrating transactions. We suggest that concepts from active object systems can support the construction of reliable event-driven applications. In particular, we are concerned with unbundling transactional reactive behavior in a CORBA environment and introduce X2TS as integration of transaction and notification services. X2TS features rich coupling modes that are configured on a per subscriber basis and supports the application programmer with coordinating asynchronous executions on behalf of transactions.

Distributed, object-oriented, active, real-time DBMSS: we want it all - do we need them (at) all?

Abstract Whenever technologies converge there exists the potential for huge benefits but also the risk of failure. The main pitfall when combining technologies that evolved independently consists in attempting to provide the union of features without properly considering the often incompatible assumptions and the crosseffects. In this paper real-time databases, active databases, and distributed object systems are analyzed together with some of the basic assumptions underlying previous work in these core technologies. Crosseffects and potential incompatibilities are discussed in an attempt to provide a better foundation for a configurable middleware platform that realistically combines selected features of active, real-time and distributed object systems.

An infrastructure for meta-auctions

Auctions have been a popular trading paradigm for centuries but have gained new interest through world-wide trading on the Internet. Many B2C sites are embracing reverse auctions as an additional service to registered customers. In all these cases the efficient notification of the participants is essential. We develop the notion of a meta-auction that allows a potential buyer to roam automatically across auction sites and we identify critical communication and notification requirements of the next generation of Internet-scale trading systems. First, todays information systems are limited in their growth and interaction potential because the typical client-server and n-tier system architectures are solely based on a request/response interaction; second, the user-initiated query metaphor from the database domain is the primary means for information acquisition; and third, many assumptions about the meaning of data and notifications provided and exchanged through the Internet are left implicit. We argue that Internet-scale business applications require publish/subscribe as an additional interaction paradigm, should leverage proactive information dissemination and caching mechanisms, and that there is a compelling need for metadata-based infrastructures providing common vocabularies for semantically meaningful exchange of data and notifications. We illustrate these points through examples from the auction domain and the development of the meta-auction concept.