With chip makers continuing to increase the number of cores they include on each new generation of their processors, perhaps it's time to rethink the basic architecture of today's operating systems, suggested Dave Probert, a kernel architect within the Windows core operating systems division at Microsoft.
The current approach to harnessing the power of multicore processors is complicated and not entirely successful, he argued. The key may not be in throwing more energy into refining techniques such as parallel programming, but rather rethinking the basic abstractions that make up the operating systems model.
Today's computers don't get enough performance out of their multicore chips, Probert said. "Why should you ever, with all this parallel hardware, ever be waiting for your computer?" he asked.
Probert made his presentation at the University of Illinois at Urbana-Champaign's Universal Parallel Computing Research Center.
Probert is on the team working on the next generation of Windows, though he said the ideas in this talk did not represent any actual work his team is doing for Microsoft. In fact, he noted that many of the other architects on the Windows kernel development team don't even agree with his views. For the talk, he set out to define what a new operating system, if designed by scratch, would look like today. He concluded it would be quite different from Windows or Unix.
Today's typical desktop computer runs multiple programs at once, playing music while the user writes an email and surfs the web, for instance. "Responsiveness really is king," he said. "This is what people want."
The problem in being responsive, he noted, is "how does the OS know [which task] is the important thing?" You don't want to wait for Microsoft Word to get started because the antivirus program chose that moment to start scanning all your files.
Most OSes have some priority scheduling to avoid these bottlenecks, but they are still crude. (Probert even suggested telemetry, in the form of a "This Sucks!" button on each computer, that a user can push whenever he or she gets frustrated with the computer's pokiness. The resulting data could be compiled to give OS developers a better idea of scheduling.)
As they began adding multiple processor cores, chip makers took a "Field of Dreams" approach to multicore chips: Build them and hope the application programmers would write programs for them. The problem is today's desktop programs don't use the multiple cores efficiently enough, Probert said. To get the full benefit from multiple cores, developers need to use parallel programming techniques. It remains a difficult discipline to master and hasn't been used much, outside of specialised scientific programs such as climate simulators.
Perhaps a better way to deal with multiple cores is to rethink the way operating systems handle these processors, Probert said. "Really, the question is, not how do we do parallel, but what do we do with all these transistors?"
The current architecture of operating systems is based on a number of different abstractions, he explained.
In the early days of computing, one program was run on a single CPU. When we wanted multiple programs to run on a single processor, the CPU time was sliced up into processes, giving each application the illusion that it was running on a dedicated CPU. The idea of the process was an abstraction, and wouldn't be the last one. Once the OS started juggling multiple programs, it needed a protected space, free from user and program interference. Thus was born the kernel mode, which is separate from the space in which the programs were run, the user mode. In effect, kernel mode and user mode abstracted the CPU into two CPUs, Probert said.
With all these virtual CPUs, however, come struggles over who gets the attention of the real CPU. The overhead of switching between all these CPUs starts to grow to the point where responsiveness suffers, especially when multiple cores are introduced.
But with Intel and AMD predicting that the core count of their products will continue to multiply, the OS community may be safe in jettisoning abstractions such as user mode and kernel mode, Probert argued. "With many-core, CPUs [could] become CPUs again," he said. "If we get enough of them, maybe we can start to hand them out" to individual programs.
In this approach, the operating system would no longer resemble the kernel mode of today's OSes, but rather act more like a hypervisor. A concept from virtualisation, a hypervisor acts as a layer between the virtual machine and the actual hardware.
The programs, or runtimes as Probert called them, themselves would take on many of the duties of resource management. The OS could assign an application a CPU and some memory, and the program itself, using metadata generated by the compiler, would best know how to use these resources.
Probert admitted that this approach would be very hard to test out, as it would require a large pool of existing applications. But the work could prove worthwhile. "There is a lot more flexibility in this model," he said.