Balancing time predictability and high-performance

Wei Zhang, Ph.D.

Time predictability is crucial in hard real-time and safety-critical systems such as aircraft, automotive and medical controls. The accurate deployment of airbags or the functionality of heart pacemakers, for example, where timing is critical, can mean life or death in many scenarios.

Fortunately, scenarios such as these are being put to the test at the Virginia Commonwealth University School of Engineering where , associate professor in the Department of Electrical and Computer Engineering, is working to perform worst-case execution time (WCET) analysis of critical real-time applications in his research lab.

Because of changes in technology and advancements in computer architecture design, modern microprocessors now achieve much higher performance than their predecessors. At the same time, they have become increasingly complex and unpredictable with respect to timing behaviors.

Historically, computer architecture research has focused on innovations to improve the average-case performance and energy efficiency, which are often harmful to the time predictability of computing.

Many emerging real-time applications require both time predictability and high performance. Speech recognition software, for example, must be computed with real-time deadlines to be useful; otherwise, users have to speak slowly. With the widespread use of portable computing devices such as smartphones and tablets, the deep integration of speech recognition and other functions such as face and motion recognition into those devices will offer tremendous benefits in terms of convenience, security, entertainment and usability among others. However, for maximum utility, these applications must be performed in real-time.

Conventional architectural design presents severe challenges when real-time guarantees are required. So the need for high-performance and time-predictable processors that can improve not only hard real-time tasks but also non hard real-time tasks becomes even more important. Balancing time predictability and performance is likely to benefit more applications, increasing the quantity of the processors with customized architectures targeted towards real-time applications which will eventually lower the manufacturing cost per chip.

The objective of Zhang’s research is to design a time-predictable yet high-performance multicore architecture to provide predictable performance for future high-performance real-time applications ranging from aircraft and automobile control systems to emerging applications for smartphones, without significantly impacting the performance and energy efficiency of other non-real-time applications.

Even when predictability isn’t as critical, as in video gaming and graphic animations, which are important parts of our society, real-time applications must be able to harness the full potential of multicore chips in a deterministic manner, Zhang says.

“Time predictability is a delicate balancing act which requires hardware and software cooperation and design trade-offs.”

The National Science Foundation, IBM, Intel, Motorola and Altera have supported Zhang’s research. He is a senior member of the Institute of Electrical and Electronics Engineers.

“For real-time applications to harness the full potential of multicore chips, the execution time of multicore processors must be predictable, which is particularly crucial for hard real-time and safety-critical systems.”

– Wei Zhang, Ph.D.