Hi Alex,
On Wed, 19 Jan 2000, Alex Khripin wrote:
> I think they make a number of very good points. Anyways, is anybody
> going to integrate their patch for the improved task structure in the
> 2.3 series? Additionally, I think they make some good points about a
> many-to-one implementation. I have some of my own thoughts on that.
> Many to one is certainly useful, but I have an idea that's a bit
> different. Perhaps a way to arrange threads in recursive clusters would
> be possible. The scheduler then goes through the top level clusters, and
> decides on each one's goodness value. Then, if the top priority option
> is a process, it simply selects it. If it's a cluster, it runs the
> goodness selection again on all it's members, and so on. This way, we
> can go from a linear relationship with time/processes to a logarithmic
> one (assuming there is a good infrastructure in place for allowing
> processes to do this) This gives the scheduling benefits of green
> threads without all the kludges.
this is exactly the job done by my June 99 patch.
It subdivides runnable tasks in priority "slots" ( clusters ) so to find the
best one You search into the higher available and if You find one You can stop
Your loop having found the _next_.
This way to handle the search gives the same speedup as the hashing do versus
linked lists ( logarithmic vs. linear ).
And the results of my tests prove this.
Cheers,
Davide.
-- "Debian, the freedom in freedom."- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sun Jan 23 2000 - 21:00:21 EST