Application Areas

Three aspects of communication have to be covered when designing an object oriented data acquisition system for a large hadron collider experiment:

1. Readout Unit - Filter Unit communication. Data are requested by processing elements. Intelligent memory devices act as servers for these requests. These servers are implemented as real-time systems and will have to handle a cumulative object request rate of 100 kHz from about 500 computing nodes.

2. Interprocess Communication. Tasks on the filter unit should be embedded into a framework for ease of management. During the lifetime of the experiment they will be exchanged. Therefore a communication abstraction that survives changes in technology (hardware platforms, operating systems, programming languages, generations of developers) has to be found. This communication layer must be lightweight, delivering performance close to what is provided by the hardware subsystems.

3. Management of the Cluster. As the cluster is heterogeneous (real-time systems, workstations, display stations) a management system is required that allows extensibility and platform independence. Management policies shall be freely definable so that the system can be tailored to changing configurations.

Our work has to reflect all three deployment areas. In the following sections I will investigate if CORBA fits the first two issues. On the third issue I will elaborate a solution in chapter 6.3.

We used TAO as a tool for evaluating a possible application of CORBA in the above mentioned areas. A lot of research has been invested by the creators of TAO to document the performance improvements that have been chosen in order to achieve scalability and efficient object invocation. We are now interested in how the ORB performs compared to use of the bare communication channel. We followed the development of the ORB over a period of one and a half year. The following sections describe evaluations that have been made to complete our picture of ORB performance. It seems legitimate to draw conclusions from these evaluations as we are always interested in the performance relative to other approaches. Thus, with better hardware and operating system support the difference will stay the same, giving us an impression of the overhead that we have to pay to get an object oriented abstraction of the communication subsystem.