diff options
| author | Florian Fischer <florian.fl.fischer@fau.de> | 2019-10-16 14:46:03 +0200 |
|---|---|---|
| committer | Florian Fischer <florian.fl.fischer@fau.de> | 2019-10-16 14:46:03 +0200 |
| commit | d22ab3d97a30780a930063b120ef3386aad8578e (patch) | |
| tree | 6b081ffeeb0a1782e19706f659f32f5db687fedb | |
| parent | 538199590fefb8814230cfc4481f63a791880ae0 (diff) | |
| download | allocbench-d22ab3d97a30780a930063b120ef3386aad8578e.tar.gz allocbench-d22ab3d97a30780a930063b120ef3386aad8578e.zip | |
increase workload size for bigger max speedup
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.
| -rw-r--r-- | src/benchmarks/falsesharing.py | 6 | ||||
| -rw-r--r-- | src/benchmarks/falsesharing/Makefile | 4 |
2 files changed, 6 insertions, 4 deletions
diff --git a/src/benchmarks/falsesharing.py b/src/benchmarks/falsesharing.py index 5debd94..a9c72b5 100644 --- a/src/benchmarks/falsesharing.py +++ b/src/benchmarks/falsesharing.py @@ -38,10 +38,10 @@ class BenchmarkFalsesharing(Benchmark): def __init__(self): name = "falsesharing" - self.cmd = "cache-{bench}{binary_suffix} {threads} 100 8 1000000" + self.cmd = "cache-{bench}{binary_suffix} {threads} 100 8 10000000" self.args = {"bench": ["thrash", "scratch"], - "threads": Benchmark.scale_threads_for_cpus(2)} + "threads": Benchmark.scale_threads_for_cpus(1)} self.requirements = ["cache-thrash", "cache-scratch"] super().__init__(name) @@ -84,12 +84,14 @@ class BenchmarkFalsesharing(Benchmark): ylabel="'l1 cache misses in %'", title="'cache misses: ' + arg + ' ' + str(arg_value)", filepostfix="l1-misses", + autoticks=False, fixed=["bench"]) self.plot_fixed_arg("({LLC-load-misses}/{LLC-loads})*100", ylabel="'llc cache misses in %'", title="'LLC misses: ' + arg + ' ' + str(arg_value)", filepostfix="llc-misses", + autoticks=False, fixed=["bench"]) diff --git a/src/benchmarks/falsesharing/Makefile b/src/benchmarks/falsesharing/Makefile index 763864a..2058b05 100644 --- a/src/benchmarks/falsesharing/Makefile +++ b/src/benchmarks/falsesharing/Makefile @@ -3,8 +3,8 @@ OBJDIR ?= obj CXX ?= g++ WARNFLAGS ?= -Wall -Wextra -COMMONFLAGS ?= -fno-builtin -fPIC -DPIC -pthread -OPTFLAGS ?= -O3 -DNDEBUG +COMMONFLAGS ?= -fno-builtin -fPIC -DPIC -pthread -g +OPTFLAGS ?= -O0 CXXFLAGS ?= $(OPTFLAGS) $(WARNFLAGS) $(COMMONFLAGS) |