Monday May 07, 2007
Monday May 07, 2007
This infrastructural upgrade has replaced the kernel's legacy CMT scheduling framework (implemented in chip.c/chip.h), which provided for only a single level of optimization (physical processor). The PG infrastructure enables Solaris to optimize for N levels, which is needed in cases where multiple levels/types of hardware sharing exists between logical CPUs in the system. Longer term, we're interested in providing a PG based implementation for the scheduling side of the kernel's NUMA optimizations. In addition to simplifying the implementation, this would potentially get us to the point of a having an affinity/load balancing policy implementation that spans both NUMA and CMT.
The road ahead is an exciting one
Over the next year or two, i'll be focusing my efforts on Solaris platform independent power management policy...which will entail bringing (or coordinating to bring) power management awareness to the platform independent kernel subsystems that deal with power manageable resources. We'll start with the dispatcher. :)
"Tesla" is the code name for the project, which will be run "in the open" via OpenSolaris. Over the last week, i've been working on the logistical aspects of the project (getting content on the page, setting up mail aliases, figuring out the Mercurial SCM, etc.). I'm hoping the project will go live either tomorrow (uh, today) or Tuesday.
Posted by Andrei Dorofeev on May 08, 2007 at 05:23 PM PDT #