… after a hot summer when I basically ignored writing on this blog.
Over the summer, my paper submitted to the Workshop on Power Aware Computing and Systems (Hot Power 2009), co-located with the Symposium on Operating Systems Principles, one of the most prestigious conferences on the subject was accepted and thus I will travel to Montana to present it. If you are interested, you can download it from the workshop web-page (soon) or leave me a message.
Essentially, I’m proposing a HW architecture and an OS level scheduling mechanism that relies on scaling the frequency at which some cores execute in a multi-core chip, according to what kind of application – single-threaded or multi-threaded – is being executed. Crucially, this scaling happens at the request of the application itself, the OS making sure that the overall power budget and the priority of applications is respected. In fact, the proposal advocates handling the speed at which different cores execute and the overall power budget of the chip as a resource – similar to memory – allocated/de-allocated based on application requests. The paper argues that near-linear speed-up may be obtained in some cases even for mostly sequential applications, ‘tricking’ in this way Amdahl’s law.
Of course, the paper has a number of shortcomings and white spots which I’ll let you discover . I’m really looking forward to this workshop in Montana and hope to see many of you there .