allocbench/src/benchmarks/falsesharing, branch master benchmark tool for POSIX memory allocators make the whole project more python idiomatic 2020-06-02T09:18:47+00:00 Florian Fischer florian.fl.fischer@fau.de 2020-05-06T14:56:32+00:00 8174a918ea3b7cb216bf7ea98cfdc10661b5c37d * rename src directory to allocbench * make global variable names UPPERCASE * format a lot of code using yapf * use lowercase ld_preload and ld_library_path as Allocator members * name expected Errors 'err' and don't raise a new Exception * disable some pylint messages 
* rename src directory to allocbench
* make global variable names UPPERCASE
* format a lot of code using yapf
* use lowercase ld_preload and ld_library_path as Allocator members
* name expected Errors 'err' and don't raise a new Exception
* disable some pylint messages
increase workload size for bigger max speedup 2019-10-16T12:46:03+00:00 Florian Fischer florian.fl.fischer@fau.de 2019-10-16T12:46:03+00:00 d22ab3d97a30780a930063b120ef3386aad8578e with the current workload size of 100K allocations the parallel part (worker) takes on my development laptop 96% of the single threaded execution time resulting in a maximal speedup of 25. (Amdahl: 1/(1-0.96) ~= 25) Increasing the the workload to 1M allocations ups the parallel portion to ~99% which gives us a max speedup of ~295 enough for all our test machines. With the new workload the single threaded execution time is ~100s on my laptop and ~30s on one of the test servers which is tolerable in my opinion. 
with the current workload size of 100K allocations the parallel part (worker)
takes on my development laptop 96% of the single threaded execution time
resulting in a maximal speedup of 25. (Amdahl: 1/(1-0.96) ~= 25)
Increasing the the workload to 1M allocations ups the parallel portion
to ~99% which gives us a max speedup of ~295 enough for all our test machines.
With the new workload the single threaded execution time is ~100s on my
laptop and ~30s on one of the test servers which is tolerable in my
opinion.
rework build system #2: call make before executing any benchmark 2019-02-02T13:45:21+00:00 Florian Fischer florian.fl.fischer@fau.de 2019-02-02T13:45:21+00:00 8c12b0494aa241bb6ddf3779a99d2f5d6ededf73 Make scans the project directory for Makefiles and executes them. The targets should be build by the new targets/Makefile. All included Makefiles are now "quiet" by default. 
Make scans the project directory for Makefiles and executes them.
The targets should be build by the new targets/Makefile.
All included Makefiles are now "quiet" by default.
rework build system #1 2019-02-01T15:35:20+00:00 Florian Fischer florian.fl.fischer@fau.de 2019-02-01T15:35:20+00:00 130765de719a3ddc475284e13749d09ff371a8e1 each benchmark has its own Makefile which must put it's binaries into OBJDIR which is added to the PATH during execution. 
each benchmark has its own Makefile which must put it's binaries into
OBJDIR which is added to the PATH during execution.