SCHED_DEADLINE at Linux Plumbers Conference '12
The Linux Plumbers Conference is a forum, with strong technical focus, that spurs communication and problem-solving for system-wide issues crossing project and business boundaries. Every year it brings together highly skilled Linux developers from all over the world setting the ground for next year advancements of the low level Linux infrastructure. This year the S(o)OS project has participated with a talk by Juri Lelli from SSSA with title “SCHED_DEADLINE (what it does and doesn't do.. yet)”.
SSSA's student Lelli, current maintainer of the SCHED_DEADLINE patchset, started giving an overview of what the patchset implements, how it interacts with the Linux scheduler, and why it would be beneficial for Linux to have it merged in the stock kernel code. Several concepts coming from the Real-Time community appear indeed still esoteric to the vast majority of Linux experts. In this regard, the talk helped also to shed some light on deadline scheduling, temporal isolation and distributed real-time load balance.
A detailed description of the current status of the patchset (i.e., what Linux version the code is based on, where it resides, how it is tested), and a list of desiderable additional features to be soon implemented, raised quite an interest in the audience. Several attendees, belonging to major high-tech companies, expressed their interest in evaluating the adoption of SCHED_DEADLINE for forthcoming products. They also wanted to know about chances for mainline inclusion. The answer came directly from Linux schedulers mainteiners that were present saying that the code itself is at a good level, and it would be ready for being merged, but they are only waiting for an industrial level use case to “sell the patchest to Linus Torvalds”.
Lelli concluded the talk with a summary on how and where SCHED_DEADLINE is currently used (both the research community and industrial entities like Ericsson or Wind River are working with it) and presenting additional fields of interest, power aware scheduling on heterogeneous multi-core architectures being a promising one.