This makes the memory source and sinks more realistic: we can read and write using a Future, so this seems like it is actually something you could potentially use in production.
Some poor-man's benchmarking shows this is a lot faster than cascading local mode:
some example timings:
scalding: 1.24314 ms
cascading: 210.13103 ms
scalding: 0.497229 ms
cascading: 205.542811 ms
scalding: 1.300902 ms
cascading: 207.588402 ms
scalding: 9.135427 ms
cascading: 248.029985 ms
scalding: 0.4009 ms
cascading: 0.750325 ms
scalding: 0.276766 ms
cascading: 0.656262 ms
scalding: 6.968218 ms
cascading: 218.509877 ms
scalding: 3.855046 ms
cascading: 214.557583 ms
Of course we can optimize more and look better use parallelism.
This makes the memory source and sinks more realistic: we can read and write using a Future, so this seems like it is actually something you could potentially use in production.
Some poor-man's benchmarking shows this is a lot faster than cascading local mode:
some example timings:
Of course we can optimize more and look better use parallelism.