The SKaMPI-Benchmark is a suite of tests designed to measure the performance
of MPI. MPI (Message Passing Interface) is a machine independent standard to
provide an efficient interface to message passing libraries of different
parallel machines. The goal of SKaMPI is to create a database to illustrate
the performance of these different MPI implementations. This database should
allow to:
1. Decide how to implement portable and efficient MPI programs.
2. Predict the performance of MPI programs.
3. Describe supercomputers in terms of MPI.
and portability stuff:
Added -z option. This is to be used with the -a option. It
provides more complete testing for small record sizes
when the file sizes are very large.
Added -y and -q to set record size range
Added command line to output
Put auto cross over back to 16 Meg
Add labels to the latency/offset output files.
Prevent mixed modes. Auto and throughput.
Added support for the Plus extended options.
Added support for -+u option. Cpu utilization.
Added network testing mode. -+m (Experimental) Tested: Linux, HP-UX
Added -xflag support for distributed mode.
Handle interrupts when in distributed mode.
Disable CPU utilization in distributed mode.
Add -+m cluster option to the help list and the list of options.
Enable more options in Cluster mode.
Add protocol version checking for distributed messages.
Add -+d file I/O diagnostic mode.
Fix some compiler warnings and implement the -+x option for
setting the multiplier used for file and record size incrementing.
Reduce the message traffic due to master's distribution of STOP. Only
one STOP distribution is needed. More can lead to socket buffer overflows.
Add -+p percentage read option.
Improve the mixed mode distribution algorithm.
Introduce -+r for O_RSYNC.
Add speed check code.
Increase maximum threads/procs to 256
Add contribs and -+t to help splash screen.
Disable fread and fwrite testing if mmap or async is in use.
Add pread/pwrite throughput testing
Permit smaller values for -n and -g
Make initial write in initfile() a page size request.
Stop test if file can not be written.
bonnie++ (1.03a) unstable; urgency=low
* Fixed the bugs in GCC 3.2 compilation and compiled with GCC 3.2. Results
will be slightly lower now, but it's still a fair test of compiler/OS...
bonnie++ (1.03) unstable; urgency=low
* Fixed a bug in error handling during a stat test with multiple directories.
bonnie++ (1.02d) unstable; urgency=low
* Fixed a bug where it would SEGV if you specified more than 999 directories
(now it handles 99,999 and displays an error message if you ask for more).
Closes: #171507
* Made it build on Irix.
* Fixed <vector> and <algo> checks to work with GCC 3.2.
have it be automatically included by bsd.pkg.mk if USE_PKGINSTALL is set
to "YES". This enforces the requirement that bsd.pkg.install.mk be
included at the end of a package Makefile. Idea suggested by Julio M.
Merino Vidal <jmmv at menta.net>.
moving the state directory for bytebench to /var/benchmarks instead of
trying to keep state in ${PREFIX}/share/bytebench. Distribution changes
from version 3.6 include:
* Double precision Whetstone instead of the old "double" benchmark.
* Removal of some obsolete files.
* "system" suite adds shell8.
* perlbench and poll added as "exhibition" (non-index) benchmarks.
* Minor change to fstime.c to fix overflow problems on fast machines.
* A lot more operating system-oriented tests into the index.
* Many tests rewritten to be more relevant for modern processors.
* New baseline: SPARCstation 20-61 with 128 MB RAM, a SPARC Storage
Array, and Solaris 2.3
pkgsrc. Instead, a new variable PKGREVISION is invented that can get
bumped independent of DISTNAME and PKGNAME.
Example #1:
DISTNAME= foo-X.Y
PKGREVISION= Z
=> PKGNAME= foo-X.YnbZ
Example #2:
DISTNAME= barthing-X.Y
PKGNAME= bar-X.Y
PKGREVISION= Z
=> PKGNAME= bar=X.YnbZ (!)
On subsequent changes, only PKGREVISION needs to be bumped, no more risk
of getting DISTNAME changed accidentally.
