How to Maximize CPU Utilization - @spawn Assigning to Busy Workers - Use pmap Instead

Specific Domains Julia at Scale
,
13

Benchmarks show no garbage collection taking place (may that be hidden?).

julia> @benchmark (@sync for n in t      | julia> @benchmark pmap(n->slow(n), t) seconds=100
           Threads.@spawn slow(n)        | 
       end) seconds=100                  | 
BenchmarkTools.Trial:                    | BenchmarkTools.Trial:
  memory estimate:  123.39 KiB           |   memory estimate:  477.23 KiB
  allocs estimate:  1378                 |   allocs estimate:  12356
  --------------                         |   --------------
  minimum time:     13.207 s (0.00% GC)  |   minimum time:     11.686 s (0.00% GC) 
  median time:      13.588 s (0.00% GC)  |   median time:      12.136 s (0.00% GC)
  mean time:        13.535 s (0.00% GC)  |   mean time:        12.075 s (0.00% GC) 
  maximum time:     13.726 s (0.00% GC)  |   maximum time:     12.427 s (0.00% GC)
  --------------                         |   --------------
  samples:          8                    |   samples:          9
  evals/sample:     1                    |   evals/sample:     1

The times here are not realistic due to thermal throttling but still show the difference.

3 Likes