In a reply to Netflix (or on handling large amounts of data efficiently in perl), I suggested that Judy::1/Judy1(3) was a sparse bit vector which might be interesting as a replacement for vec(). Perl's built-in vec can also be used as a bit vector but it's not sparse - all bits between the zeroth to the highest ever set are instantiated inside a perl string. This is only convenient if your bit vector isn't too big in ram.
Example code
A simple perl bit vector
use constant MEGABYTE => 2 ** 20;
my $vector = '';
# 5,000 bits. Set every 100th bit between 15 million and 20 million.
for ( 15_000_000 .. 20_000_000 ) {
if ( ! ( $_ % 100 ) ) {
vec( $vector, $_, 1 ) = 1;
}
}
printf "%0.1fM\n", length( $vector ) / MEGABYTE;
A simple Judy::1 bit vector
use constant KILOBYTE => 2 ** 10;
use Judy::1 qw( Set MemUsed );
my $vec;
# 5,000 bits. Set every 100th bit between 15 million and 20 million.
for ( 15_000_000 .. 20_000_000 ) {
if ( ! ( $_ % 100 ) ) {
Set( $judy, $_ );
}
}
printf "%0.1fK\n", MemUsed( $judy ) / KILOBYTE;
Memory "benchmarks"
Density | vec() | Judy::1 |
Tests using the entire range 0..20,000,000 |
Every bit | 2.3M | 618K |
Density | vec() | Judy::1 |
Tests using range 15,000,000..20,000,000 |
Every bit | 2.3M | 162K |
Every 10th bit | 2.3M | 772K |
Every 100th bit | 2.3M | 158K |
Every 1000th bit | 2.3M | 65K |