Here is my LinuxCon 2011 presentation.
Absract:
In an ideal scenario scheduler should be adaptable to the workload characteristics and provide best performance so that system administrators or users need not be experts to tune it. The Linux scheduler (CFS – Completely Fair Scheduler) is found to scale in various environments, right from embedded systems to large servers. It has got more than ten tunables which can significantly influence functionality and performance, however their impact on various workloads have not been sufficiently studied. This presentation is an attempt to educate on the scheduler tunables and how they can be carefully tuned to influence performance and functionality of different types of workloads such as CPU intensive, IO intensive, etc. This talk would be beneficial for those who like to understand how to control the behaviour of the scheduler. People with intermediate level of knowledge in Linux environment can take part in this discussion.
Link to the presentation: http://events.linuxfoundation.org/slides/2011/linuxcon/lcna2011_rajan.pdf
