|| ||Clark Williams <firstname.lastname@example.org> |
|| ||RT <email@example.com> |
|| ||[rt-tests] version 0.63 available |
|| ||Wed, 27 Jan 2010 10:23:54 -0600|
|| ||LKML <firstname.lastname@example.org>, Carsten Emde <email@example.com>,
<firstname.lastname@example.org>, John Kacur <email@example.com>,
Thomas Gleixner <firstname.lastname@example.org>|
|| ||Article, Thread
The latest version of rt-tests is now available as a tarball from:
Also available in my git repo:
under tag v0.63
This version includes a new option: --numa, which is a counterpart to
the recently introduced --smp option.
The --smp option is a shorthand for: -t -a -n, meaning that one
measurement thread is created per active CPU and is affined to that cpu
and that clock_nanosleep is used for measurements.
The --numa option does the same options as --smp, but also makes calls
into libnuma to ensure that measurement threads are always bound to the
memory node that is local to it's cpu (avoiding cross-node memory
references). Each measurement thread's stack and major data structures
are allocated from the local memory node.
Note that to actually use --numa mode, you must pass NUMA=1 to the make
$ make NUMA=1
This is for all you embedded guys that could care less about NUMA :)
Other changes from previous releases are that the hwlatdetect test has
been moved to the python site-library (so that other tests can import
it) and then symlinked to /usr/bin/hwlatdetect. Also, copyright
messages have been updated in the source files (thanks Uwe).