Technical Report Number
General purpose middleware has been shown effective in meeting diverse functional requirements for a wide range of distributed systems. Advanced middleware projects have also supported single quality-of-service dimensions such as real-time, fault tolerance, or small memory foot-print. However, there is limited experience supporting multiple quality-of-service dimensions in middleware to meet the needs of special purpose applications. Even though general purpose middleware can cover an entire spectrum of functionality by supporting the union of all features required by each application, this approach breaks down for distributed real-time and embedded sys-tems. For example, the breadth of features supported may interfere with small memory footprint requirements. In this paper, we describe experiments comparing application-level and mechanism-level real-time perfor-\mance of a representative sensor-network application running on three middleware alternatives: (1) a real-time object request broker (ORB) for small-footprint networked embedded sensor nodes, that we have named nORB, (2) TAO, a robust and widely-used general-purpose Real-Time CORBA ORB, and (3) ACE, the low-level middleware framework upon which both nORB and TAO are based. This paper makes two main contributions to the state of the art in customized middleware for distributed real-time and embedded applications. First, we present mechanism-level timing measurements for each of the alternative middleware layers and compare them to the observed performance of the sensor-network application. Second, we provide a preliminary performance model for the observed application timing behavior based on the mechanism-level measurements in each case, and suggest further potential performance optimizations that we plan to study as future work.
Subramonian, Venkita; Xing, Guoliang; Gill, Christopher; and Cytron, Ron, "Towards a Performance Model for Special Purpose ORB Middleware" Report Number: WUCSE-2003-5 (2003). All Computer Science and Engineering Research.