Chunking complements event-driven programming. For example reaching out to millions of devices.

use strict;
use warnings;

use MCE::Flow;
use MCE::Queue;
use AnyEvent::SNMP;

my $work_q = MCE::Queue->new;
my $nthreads = 100;
my $chunk_size = 300;

sub consumer {
    while ( defined ( my $data = $work_q->dequeue ) ) {
        MCE->printf("wid %3d, size %d\n", MCE->wid, scalar(@$data));

        MCE->yield; # Optional
            # The delay occurs serially among workers.
            # It is helpful to not all blast simultaneously.

        # do event-driven programming to 300 max items
        # ...
    }
}

sub provider {
    my @input = (1 .. 20_000);
    while (@input) {
        my @data = splice @input, 0, $chunk_size;
        $work_q->enqueue(\@data);
    }
    $work_q->end;
}

MCE::Flow->run(
    # 100 consumers, 1 provider
    { max_workers => [ $nthreads, 1 ],
      interval => {delay => 0.008} },
    \&consumer, \&provider,
);
[download]

It depends on the app. I tend to set nthreads to about 4x the number of logical cores on the box. The delay is serially behind the scene. Each worker waits it's turn in milliseconds before initiating to the OS eventIO to hundreds of devices.