Performance

Performance measurements were taken using std::chrono::highresolution_clock, with overhead corrections. The code was compiled using the build options: variant = release, optimization = speed ^[1].

Table 1.1. Overhead of creating and joining

thread	fiber	tbb	qthread
31 µs	1.1 µs	570 ns	620 ns

Table 1.2. Overhead of detach

thread	fiber
20 µs	3.2 µs

Table 1.3. Overhead of yield

thread	fiber
38 µs	1.3 µs

Table 1.4. Overhead of waiting on a future

thread	fiber
32 µs	3.0 µs

Table 1.5. Scaling of creating and joining

average of	thread	fiber
10	50.65 µs	4.83 µs
50	52.99 µs	4.84 µs
100	50.44 µs	5.24 µs
500	45.19 µs	4.86 µs
1000	42.59 µs	5.04 µs
5000	42.30 µs	5.07 µs
10000	41.07 µs	5.12 µs

Using internally atomics by applying BOOST_FIBER_NO_ATOMICS.

Table 1.6. Overhead of creating and joining

thread	fiber	tbb	qthread
31 µs	955 ns	570 ns	620 ns

Table 1.7. Overhead of detach

thread	fiber
20 µs	3.2 µs

Table 1.8. Overhead of yield

thread	fiber
38 µs	1.1 µs

Table 1.9. Overhead of waiting on a future

thread	fiber
32 µs	2.4 µs

Table 1.10. Scaling of creating and joining

average of	thread	fiber
10	50.65 µs	3.76 µs
50	52.99 µs	2.78 µs
100	50.44 µs	2.45 µs
500	45.19 µs	2.91 µs
1000	42.59 µs	3.60 µs
5000	42.30 µs	4.57 µs
10000	41.07 µs	4.21 µs

^[1] Intel Core2 Q6700, x86_64, 3GHz