Re: Make yield_task_fair more efficient

[Balbir Singh <balbir@linux.vnet.ibm.com>]

Ingo Molnar wrote:
> * Balbir Singh <balbir@linux.vnet.ibm.com> wrote:
> 

You did not answer some of my earlier questions.

>> I have an alternate approach in mind (that I need to find time for), 
>> threaded-rbtrees. Walking the tree is really efficient, specially 
>> finding successor of a node.
> 
> sure, feel free to experiment with those details.
> 
> But if you want to improve Java workloads then you should probably start 
> by converting them to futexes instead of sched_yield(), that will 
> probably give far more performance than micro-optimizing the 
> sys_sched_yield() codepath. (especially when it happens at the expense 
> of other workloads, which is not acceptable for mainline) You can 
> rebuild your JVM easily and re-test with its locking fixed, right?

No.. I don't want to optimize the JVM or rebuild it. I don't have access to any
JVM code either. I wanted to do the threaded rb-trees for the case where we use
spend time finding the successor using rb_next() (walking the tree).

-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL