[openbeos] Re: GSoC ticket #1069 (thread scheduler) application
- From: Simon Taylor <simontaylor1@xxxxxxxxxxxx>
- To: <openbeos@xxxxxxxxxxxxx>
- Date: Mon, 9 Apr 2007 11:06:15 +0000
> From: "André Braga" <meianoite@xxxxxxxxx>
>
> Pursuing the perfect algorithm for all given workloads on an
> uniprocessor system is a quixotesque task, IMVHO. There are better
> windmills to fight out there. Besides, Haiku has a clear goal of
> desktop responsiveness, so that's the kind of tuning we're interested
> in.
Amen to that (not that I've ever heard of fighting windmills before!)
> That said, I would definitely like to experiment with *offline*
> optimization of tunables using GAs once I'm set with an algorithm of
> my liking; I even hinted at that on my GSoC application form. But I
> can't really subscribe to bolting a GA runtime mutation engine onto
> the Haiku kernel. IMO the overhead is just not worth it.
Absolutely agreed. Offline optimisation using a set of standard desktop use
cases sounds sensible but doing it at runtime is silly. Doesn't really fit with
the lean, fast, small-kernelled approach of Haiku. The other problem I forsee
is that it would be hard for application developers to create applications with
good threading policies (relative priorities and stuff) if the OS is going to
go changing its scheduler on you at run-time.
> Cheers,
> A.
Simon
-----------------------------------------
Email sent from www.virginmedia.com/email
Virus-checked using McAfee(R) Software and scanned for spam
Other related posts:
