I had a chance to pose the following questions to Kelvin Nilsen, Atego Chief Technology Officer, about Real Time Java and Atego's various implementations. Also check out my hands-on review of Atego's AonixPERC Ultra SMP real time Java implementation and an earlier video interview with Atego's Gary Cato.
Wong: Where does Atego see real time Java fitting into developer's plans?
Kelvin: We see several trends with respect to real-time systems in general. First, modern microprocessor designs (with large multi-level memory caches, deep instruction pipelines, speculative and out-of-order instruction execution) make it much more difficult to do the rigorous timing analysis required by hard real-time protocols. Thus, there is a trend to rearchitect hard real-time systems to use soft real-time development approaches. Second, many real-time systems track Moore's law, doubling in size every 18 to 36 months.
Many of today's real-time systems are comprised of multiple millions of lines of code. In previous decades, it was common for two or three developers to develop all of the software required for each new embedded system within 6-12 months of development time. Today's systems are so large that this approach is not possible. These systems are implemented largely by assembling existing off-the-shelf software components. The typical transition towards the use of real-time Java proceeds incrementally. As system requirements evolve, new capabilities must be added to existing embedded systems. As development managers search for existing off-the-shelf components to address the new requirements, they are much more likely to find the desired capabilities as reusable Java components. If they cannot find the desired capabilities as off-the-shelf components, then the company will have to develop the capability from scratch.
The choice to use Java for new development is motivated by several factors. First, doing development in Java is typically twice as productive as doing it in C or C++. Second, the anticipated long-term software maintenance costs will be lower (by up to ten fold) if the software is implemented in the Java language. Finally, Java software components are generally more amenable to future modifications and generalizations in the face of possible future evolution of requirements.
Wong: What products do you have that address real time Java?
Kelvin:
- PERC Ultra best used for applications that need the rich features of Java Standard Edition (JSE) libraries, coupled with predictable performance. AonixPERC is a mature, proven solution having been field tested in more applications than any other real-time virtual machine system. It has demonstrated 5-9s reliability in applications running literally for over one year with no restarts.
- PERC Pico is specifically designed for demanding applications requiring small footprint, fast execution, and low-level device access. Benchmarks have shown Pico applications perform comparably to C/C++ programs, while being fully predictable and memory efficient.
- PERC Raven profits from Atego's long tradition providing safety-critical technologies used in many safety-intense applications, including avionics, space, and transportation systems where lives are on the line. AonixPERC Raven applies safety constraints, the basis for fulfilling safety certification requirements.