Subject: Re: 1:1 threading changes
To: None <firstname.lastname@example.org>
From: Christos Zoulas <email@example.com>
Date: 01/16/2007 18:08:09
In article <20070116074910.GC7524@hairylemon.org>,
Andrew Doran <firstname.lastname@example.org> wrote:
>I have committed the remaining parts of the 1:1 threading mods that I have
>been working on. Please let me know if I have managed to screw anything up,
>or if you notice any problems. This shouldn't affect system behavour because
>SA threads are the default.
>I think (hope) I have fixed the remaining problems with it. It passes all
>the regression tests, and the stress tests that I have tried. It seems to
>run a variety of apps OK, e.g.: Firefox, Java and OpenOffice.
>The main visible benefit to this is that on multiprocessor systems, runnable
>threads will be distributed among all of the CPUs. If you are interested in
>trying it out, the necessary steps are:
>o Get latest -current from CVS.
>o Edit lib/libpthread/Makefile, and comment out the 'PTHREAD_SA=1' line.
>o Build and install the system (or just libc and libpthread if you prefer).
>o Check out src/sys from the newlock2 branch, build and install a kernel.
>As noted, unfortunatley it requires a kernel from the newlock2 branch. That
>currently has a number of problems, including:
>o Only i386, amd64 and sparc64 are tested. MIPS, alpha and sparc might work.
>o Sporadic crashes have been seen on amd64.
>o Attempting to use scheduler activiations will currently cause a panic.
Whee! I should try this on my newly built jdk!