"We’ve already developed several technologies for performance engineering, and I’m looking forward to both developing those more and seeing them have actual impact in the real world. I think that’s something that CSAIL is generally famous for: We not only do things theoretically or imagine big things, but we actually do them in ways that have real impact on real people," Charles E. Leiserson.
The end is near…of Moore’s law, that is. Traditionally, we have depended on Moore’s law to drive continuous innovation in computing, but the trend must eventually end due to the laws of physics. “Moore’s law, which is the economic and technical trend by which integrated circuits got faster and faster every year due to miniaturization, is in the process of bottoming out,” says Professor Charles E. Leiserson of MIT CSAIL.
That’s because semiconductor hardware can only get so small. So far, we do not yet have the capabilities that could make circuit wires smaller than atoms. As an alternative solution, Prof. Leiserson, whose work centers on algorithms and performance engineering, is creating fast code.“With the end of Moore’s Law,” he says, “software is a major place that we can expand the capability of computing systems.”
Currently, software tends to be slow, but Prof. Leiserson sees the potential for building much faster systems with great functionality that are open to developers. “We are in the circumstance where now, increasingly, software developers will have to worry about performance themselves,” he says. “Our strategy for that is both to learn how to make code that runs fast, that’s well-adapted to the architectures that the hardware vendors provide for us, as well as to develop technologies which do not have a lot of bloat, and wear and tools to make that easy to design those kinds of software systems.”
Prof. Leiserson says that one of the greatest challenges facing the research is legacy software. “People designed legacy software without regard to the features of modern architectures such as parallelism (that is, having multiple processing cores in a processor) and things like memory hierarchy (having fast memory close to the processor and slow memory farther away).” He explains that with legacy software, since you aren’t building new software that’s faster, “you kind of have to do surgery on those applications.”
Developing faster software systems now is critical as Moore’s law comes to an end, since software is used by all industries. “The impact is being felt by leading-edge computing companies — for example, people who are developing technologies for the cloud, and web services and so forth, but increasingly, it will be a client-side application software,” Prof. Leiserson says.
As for future impact? “Leading companies will do things over the next five years, and then in the next five to ten years it’ll start to become a big issue, and ten years in all, it’s going to be: If you’re not doing it, you’ve already lost.”