Using principle patterns to optimize real-time ORBs

Evaluating policies and mechanisms to support distributed real‐time applications with CORBA

“…Thus, a compliant implementation may choose to separate threads that perform all the I/O, parse the request to identify the target POA and priority, and hand off the request to the appropriate thread in the POA thread pool, as shown in Figure 10. Such an implementation can increase average and worst-case latency and create opportunities for unbounded priority inversions [15], however. For instance, even under a light load, the server ORB incurs a dynamic memory allocation, multiple synchronization operations, and a context switch to pass a request between a network I/O thread and a POA thread.…”

“…For instance, object references created in one ORB could be used only in the thread of that ORB. Therefore, application developers could not pass object references between threads on the same CPU board because client/servant collocation did not work transparently [15].…”

“…An increasing number of research efforts are focusing on integrating QoS and real-time scheduling into middleware like CORBA. Our previous work on TAO has examined many dimensions of ORB middleware design, including static [5] and dynamic [28] operation scheduling, event processing [6], I/O subsystem [31] and pluggable protocol [32] integration, synchronous [22] and asynchronous [21] ORB Core architectures, IDL compiler features [38] and optimizations [1], systematic benchmarking of multiple ORBs [39], patterns for ORB extensibility [14] and ORB performance [15]. In this section, we compare our work on TAO with related QoS middleware integration research.…”

Collocation optimizations in an aspect-oriented middleware system

“…More recently, collocation has been incorporated in many CORBA implementations. TAO (Schmidt and Cleeland, 1999) is a high-performance and real-time CORBA system whose implementation has been guided by a set of optimization principle patterns (Pyarali et al, 1999). Essentially, such patterns are applied in TAO to increase the performance, scalability and predictability of distributed systems built using the middleware.…”

Design and Performance of Asynchronous Method Handling for CORBA

“…Research directions. Many of TAO's existing ORB optimizations [8] are based on the assumption that a single activation record handles a request. Associating a request with an activation record has certain advantages, e.g., it is faster than allocating on the heap and it is easy to analyze the request's lifetime.…”