Closed NHDaly closed 2 years ago
Here is the result from running the benchmark on my machine:
------ JULIA VERSIONINFO ------
Julia Version 1.7.3-pre.8
Commit a690e381c0* (2022-04-05 13:49 UTC)
Platform Info:
OS: macOS (x86_64-apple-darwin19.6.0)
CPU: Intel(R) Core(TM) i9-8950HK CPU @ 2.90GHz
WORD_SIZE: 64
LIBM: libopenlibm
LLVM: libLLVM-12.0.1 (ORCJIT, skylake)
Environment:
JULIA_NUM_THREADS = 4
JULIA_LOAD_PATH = @:/var/folders/fd/hymwsmhj1zd27lwlftv_yb2r0000gn/T/jl_M3CJZF
------ JULIA CPU INFO ------
Sys.CPU_THREADS = 12
Intel(R) Core(TM) i9-8950HK CPU @ 2.90GHz:
speed user nice sys idle irq
#1-12 2900 MHz 2997967 s 0 s 1241982 s 33823807 s 0 s
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Running benchmark for /Users/nathandaly/.julia/dev/MultithreadingBenchmarks/src/../bench/all_tasks_compiling.jl
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
------ TEST PARAMETERS ------
nthreads_to_test = [1, 3, 6, 9, 12]
nqueries = 5
num_ops = 10
warmup
NUM_QUERIES: 1
WORK_SIZE: 10
5 threads:
0.019208049 secs
6 threads:
0.019770852 secs
Results (omitted printing latencies column):
2Γ6 DataFrame
Row β nthreads time_secs allocs memory gctime_secs thread_counts
β Int64 Float64 Int64 Int64 Float64 Vector{Int64}
ββββββΌββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
1 β 5 0.019208 8644 540364 0.0 [1, 0, 0, 0, 0]
2 β 6 0.0197709 8644 540396 0.0 [0, 1, 0, 0, 0, 0]
Processed:
2Γ6 DataFrame
Row β nthreads time_secs speedup_factor marginal_speedup utilization latency_quantiles_ms
β Int64 Float64 Float64 Float64 Float64 Tupleβ¦
ββββββΌβββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
1 β 5 0.019208 1.0 1.0 0.2 (0.1=>23.0107, 0.5=>23.0107, 0.9β¦
2 β 6 0.0197709 0.971534 -0.0284663 0.161922 (0.1=>26.9172, 0.5=>26.9172, 0.9β¦
run benchmark
NUM_QUERIES: 5
WORK_SIZE: 10
1 threads:
0.111896811 secs
3 threads:
0.09173288 secs
6 threads:
0.098779298 secs
9 threads:
0.098101715 secs
12 threads:
0.097688362 secs
Results (omitted printing latencies column):
5Γ6 DataFrame
Row β nthreads time_secs allocs memory gctime_secs thread_counts
β Int64 Float64 Int64 Int64 Float64 Vector{Int64}
ββββββΌββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
1 β 1 0.111897 43147 2698108 0.0 [5]
2 β 3 0.0917329 43152 2698300 0.0 [2, 1, 2]
3 β 6 0.0987793 43152 2698364 0.0 [1, 1, 1, 1, 1, 0]
4 β 9 0.0981017 43152 2698396 0.0 [1, 1, 1, 1, 1, 0, 0, 0, 0]
5 β 12 0.0976884 43152 2698460 0.0 [1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0β¦
Processed:
5Γ6 DataFrame
Row β nthreads time_secs speedup_factor marginal_speedup utilization latency_quantiles_ms
β Int64 Float64 Float64 Float64 Float64 Tupleβ¦
ββββββΌβββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
1 β 1 0.111897 1.0 1.0 1.0 (0.1=>29.9244, 0.5=>31.3406, 0.9β¦
2 β 3 0.0917329 1.21981 0.109906 0.406604 (0.1=>44.1762, 0.5=>45.2292, 0.9β¦
3 β 6 0.0987793 1.1328 -0.0290051 0.188799 (0.1=>83.2134, 0.5=>92.4317, 0.9β¦
4 β 9 0.0981017 1.14062 0.00260805 0.126736 (0.1=>84.305, 0.5=>94.1237, 0.9=β¦
5 β 12 0.0976884 1.14545 0.00160879 0.0954539 (0.1=>83.5484, 0.5=>93.8798, 0.9β¦
plot results
-- saving figures in /Users/nathandaly/.julia/dev/MultithreadingBenchmarks/test --
Testing MultithreadingBenchmarks tests passed
As expected, the time remains roughly constant with increasing number of threads, since julia currently maintains a global lock around compilation. π
Looks fine to me, and it's a useful test for us. π
BTW, may want to clean up this.
Also BTW, it'd be interesting to see what happens to these compiling tasks when GC is triggered...
agreed on both counts! but i don't really have the bandwidth now to maintain this package any more than this. :'(
Reenable the benchmark now that https://github.com/JuliaLang/julia/issues/33183 is fixed. π Haha it's been fixed since like julia 1.4 or 1.5 or something, but i haven't looked at these benchmarks in a long time! thanks for running them again, @d-netto!
Can you review this PR? I can't add you as a reviewer yet because you haven't contributed.
Also, CC: @kpamnany - can you review as well?