Rafael Vanoni wrote: > I've read a good deal of code involving lpl's but I'd like to clear a > couple of ideas around it. If anyone could comment on these, I'd > appreciate it. > > (a) lpl's are used to determine the load average of lgroups, which is > used by the dispatcher to load balance threads around > Yes, that's true. An lpl actually represents an intersection between an lgrp and a cpu partition. This subset of CPUs is what the dispatcher is most interested in...since a thread's CPU partition (processor set) constrains where it can run...and the dispatcher wants to consider the lgrp's CPUs that are within that constrained set.
> (b) each lgrp has a lpl, which in turn may have leaf lpl's for each cpu > in that lgrp > There may be multiple lpls associated with an lgrp...depending on how many cpu partitions / processor sets intersect the lgrp. The number of lpls an lgrp can have associated is limited only by the number of logical CPUs it contains. Leaf lgrps will have associated leaf lpls. Just as the lpl is the result of intersecting an lgrp with a cpu partition, the lpl hierarchy is the result of intersecting a cpu partition with the lgrp hierarchy. > (c) lpl topology is created by lgrp code > Yes...and triggered by processor set instantiation / destruction (and by creation of the default CPU partition, of course). > (d) only leaf lpl's have CPUs > In the implementation this is true (an artifact of the CPUs living on the lpl via a linked list, forcing higher order lpls to track CPUs using a composite of leaf lpls)...but conceptually this isn't true. When lpl's are re-implemented using the processor group abstraction, this confusing little discrepancy will go away. > (e) lpl's are the intersection of lgroups and CPU partitions > I should have read through a-e before answering the above. :) > I'm trying to understand where lpl's fir in the lgrp/cpus picture and > what are their goal besides keep load averages (if any). > Cool. Thanks! -Eric _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
