Multithreaded Programming
Mutexes implementations
All 10 implementations can be found in the ./students dir.
2 of them are the reference ones
8 others are my descriptively named implementations
- spinlock with yield on every attempt
- spinlock with 8 attempts to fast-forward and then yield
- spinlock with 32 attempts to fast-forward and then yield
- spinlock without yield
- ticket lock with yield on every attempt
- ticket lock with 8 attempts to fast-forward and then yield
- ticket lock with 32 attempts to fast-forward and then yield
- ticket lock without yield
Benchmark source
The benchmark was taken from pooh64 repository and slightly modified for a nicer output
Benchmark results for my implementations
By running bash bench.sh on my 8 core CPU with Ubuntu 18.04.4 LTS installed I've got the following output:
bench/inc.c
n_threads: -> 1 -> 2 -> 4 -> 8
a37h_spinlock_without_yield: 1.12; 6.98; 12.51; 8.15;
a37h_spinlock_with_yield: 1.20; 1.55; 2.57; 5.15;
a37h_spinlock_with_yield_optimized_32: 1.21; 7.21; 11.45; 7.36;
a37h_spinlock_with_yield_optimized_8: 1.75; 2.37; 4.07; 5.07;
a37h_ticket_without_yield: 1.14; 7.88; 10.77; 14.24;
a37h_ticket_with_yield: 1.38; 31.26; 36.00; 89.37;
a37h_ticket_with_yield_optimized_32: 1.50; 26.67; 37.25; 94.16;
a37h_ticket_with_yield_optimized_8: 1.43; 23.61; 35.44; 72.96;
pthread_mutex: 1.09; 4.55; 6.01; 8.98;
pthread_spinlock: 0.56; 1.37; 3.57; 4.04;
bench/io.c
n_threads: -> 1 -> 2 -> 4 -> 8 -> 16 -> 32
a37h_spinlock_without_yield: 0.51; 0.46; 0.45; 0.48; 14.29; 51.08;
a37h_spinlock_with_yield: 0.52; 0.45; 0.45; 0.51; 0.51; 0.50;
a37h_spinlock_with_yield_optimized_32: 0.52; 0.46; 0.47; 0.55; 0.54; 0.56;
a37h_spinlock_with_yield_optimized_8: 0.51; 0.45; 0.45; 0.56; 0.58; 0.64;
a37h_ticket_without_yield: 0.57; 0.45; 0.47; 0.66; 78.60; 117.67;
a37h_ticket_with_yield: 0.51; 0.48; 0.45; 0.52; 0.54; 0.66;
a37h_ticket_with_yield_optimized_32: 0.51; 0.49; 0.45; 0.49; 0.59; 0.58;
a37h_ticket_with_yield_optimized_8: 0.51; 0.50; 0.45; 0.51; 0.54; 0.57;
pthread_mutex: 0.51; 0.57; 0.48; 0.46; 0.46; 0.46;
pthread_spinlock: 0.53; 0.45; 0.44; 0.48; 0.74; 1.15;
bench/ring.c
n_threads: -> 1 -> 2 -> 3 -> 4 -> 5 -> 6 -> 7 -> 8
a37h_spinlock_without_yield: 0.31; 1.22; 1.58; 1.72; 1.68; 1.95; 2.02; 1.98;
a37h_spinlock_with_yield: 0.32; 1.09; 1.63; 1.77; 1.75; 1.91; 2.09; 2.36;
a37h_spinlock_with_yield_optimized_32: 0.32; 1.17; 1.51; 1.65; 1.69; 1.87; 2.03; 2.65;
a37h_spinlock_with_yield_optimized_8: 0.33; 1.20; 1.56; 1.70; 1.87; 1.87; 1.90; 2.21;
a37h_ticket_without_yield: 0.31; 1.17; 1.67; 1.80; 1.92; 2.25; 2.34; 2.91;
a37h_ticket_with_yield: 0.43; 1.40; 2.21; 2.21; 2.51; 2.60; 2.67; 2.61;
a37h_ticket_with_yield_optimized_32: 0.33; 1.12; 1.68; 1.84; 1.96; 2.17; 2.30; 2.61;
a37h_ticket_with_yield_optimized_8: 0.33; 1.24; 1.76; 1.87; 1.90; 2.14; 2.59; 2.53;
pthread_mutex: 0.30; 1.32; 1.62; 1.90; 1.98; 2.20; 2.32; 2.69;
pthread_spinlock: 0.23; 0.86; 1.18; 1.47; 1.50; 1.58; 1.74; 2.14;
Benchmark results visualised
Open a bench/inc.c full-sized result image
