Course of Raku / Functional, concurrent, reactive, and web programming / Concurrent programming / Hyper and race 🆕

Tuning with :batch and :degree

Both .hyper and .race accept two adverbs that control how the work is split across threads: :degree and :batch. You rarely need them, but they are what you reach for when you are tuning for speed.

:degree — how many workers

:degree sets how many worker threads run at the same time:

say (1..5).hyper(:degree(2)).map(*²); # (1 4 9 16 25)

By default the degree is close to the number of CPU cores the machine has — on an eight-core machine it is 7, leaving a core for the main program. Raising it much above the core count rarely helps, since there are no more cores to run on, and the extra coordination can even slow things down.

:batch — how many elements at a time

:batch sets how many elements each worker takes in one go before coming back for more:

say (1..6).hyper(:batch(2)).map(*²); # (1 4 9 16 25 36)

The default batch is 64, and the size is a trade-off:

• A large batch means less back-and-forth between the workers and the scheduler — less overhead, which is good when every element costs about the same.
• A small batch spreads the work more evenly when some elements are far more expensive than others, so no single worker gets stuck with all the slow ones — at the cost of more coordination.

You can set both at once, on .race just as on .hyper:

say (1..6).race(:degree(2), :batch(3)).map(*²).sum; # 91

my $h := (1..10).hyper;
say $h.^attributes.first(*.name.contains('config')).get_value($h).raku;
# HyperConfiguration.new(batch => 64, degree => 7)

They are still only hints

As with .hyper and .race themselves, these adverbs request a particular split rather than guarantee it. A compiler is free to honour them, adjust them, or ignore them entirely — so treat :batch and :degree as knobs to try when tuning, not as promises about exactly how the work will run.

Course navigation

← Unordered parallelism with race   |   Quiz — batch and degree →

💪 Or jump directly to the exercises in this section.