Have you ever wondered exactly how much memory your data structure are taking up? Normally this isn't the kind of thing that you need to worry about in Perl (Perl is optimised to use more memory so that it can use less CPU time where possible) and Perl handles all the nastiness of allocating and using memory but occasionally, just occasionally, you find yourself in a situation where you need to know this kind of thing.
The thing about memory is that when you run out, well, you run out. As soon as you start having to swap to disk with virtual memory, everything gets extremely slow. Even before you get to this point, it's worth noting the less memory your code has to deal with that for many subtle reasons the faster your code will run.
One situation that's really critical to keep track of memory usage is when you're using mod_perl. mod_perl, the system of embedding perl directly into a webserver, speeds things up a lot by amongst other things keeping a copy of global variables between requests for pages. This means that your scripts don't have to start from scratch each time at the cost of more memory usage. One of the major drawbacks is that if you're not careful then this memory usage can be multiplied over each separate Apache process - typically thirty-fold or so. Devel::Size can help identify situations where this might be a problem, and where you should take steps to ensure this data is placed in memory shared between the processes.
Okay, as an example, let's have a look at some data structures and see how much space they take up. Bear in mind that these figures are valid for my machine only (Debian Linux i386 unstable, with Debian's 5.8.0 threaded perl built with gcc 2.95.4) and the results will vary with your hardware and architecture.
Okay, let's get a list of all the files in the current directory and store it in a data structure:
#!/usr/bin/perl
# turn on perl's safety features use strict; use warnings;
# use Devel::Size, and import the total_size function use Devel::Size qw(total_size);
# use the cwd function to get the current working directory use Cwd;
# open the current directory list opendir DIR, cwd or die "Couldn't open the current directory";
# get all the files my @files; push @files, $_ while (readdir DIR); closedir DIR;
print "There are ".@files." files in your current dir\n";
print "Storing this array took up ".
total_size(\@files)." bytes\n";
If I run this script from within my "/etc" dir I get the following printed out on screen.
There are 240 files in your current dir Storing this array took up 9055 bytes
Let's try storing some information on the data. Let's pretend I want to know the modification time, the size and the filemode of each of these files.
One approach, the first that comes to mind, is to create a big hash that contains for each file in the directory a smaller hash that has the keys "mtime", "size" and "mode" which have the values for the modification time, size and filemode of the file stored in the values respectively.
# load File::stat so that 'stat()' will now return a "File::stat" # object that methods like '$st->mode' can be called on use File::stat;
# build a big hash that is keyed by the name of the file and in # which each entry points to another hash that has the size, mtime # and mode stored in it
my %stats;
foreach my $file (@files)
{
# stat the file working out when it was last modified, etc
my $stat = stat($file);
# store the mode, modification time and size of the file
# in a hash so we can access it later
$stats{ $file }{size} = $stat->size;
$stats{ $file }{mtime} = $stat->mtime;
$stats{ $file }{mode} = $stat->mtime;
}
print "Storing the the hash takes ".
total_size(\%stats)." bytes\n";
Which then prints
Storing the the hash takes 66679 bytes
Wow! That's 65KB. Now that doesn't actually sound like a lot, but if I do that kind of thing in each of my mod_perl children after they've forked, and I'm running twenty servers, then that memory usage is potentially multiplied twenty-fold. that's over a megabyte of memory right there. Let's have a look at storing it in another format...How about if I use a small array for each file instead of each of the small hashes?
# define constants that refer to the index the elements # are in the array use constant FILE_SIZE => 0; use constant FILE_MTIME => 1; use constant FILE_MODE => 2;
my %stats2;
foreach my $file (@files)
{
# stat the file working out when it was last modified, etc
my $stat = stat($file);
# store the mode, modification time and size of the file
# in an array so we can access it later
$stats2{ $file }[FILE_SIZE] = $stat->size;
$stats2{ $file }[FILE_MTIME] = $stat->mtime;
$stats2{ $file }[FILE_MODE] = $stat->mtime;
}
print "Storing the the hash takes ".
total_size(\%stats2)." bytes\n";
Which on my system now prints out:
Storing the the hash takes 41239 bytes
Which is a lot better. How about if instead of a hash of lists I use a list of hashes?
my @stats3;
foreach my $file (@files)
{
# stat the file working out when it was last modified, etc
my $stat = stat($file);
# store the mode, modification time and size of the file
# in an array so we can access it later
$stats3[FILE_SIZE]{ $file } = $stat->size;
$stats3[FILE_MTIME]{ $file } = $stat->mtime;
$stats3[FILE_MODE]{ $file } = $stat->mtime;
}
print "Storing the the array takes ".
total_size(\@stats3)." bytes\n";
Which now prints:
Storing the the array takes 38393 bytes
What's interesting here is essentially as we're saving more and more memory we're making our code less and less readable (and hence, less and less maintainable,) or to put it another way The more we try to save memory the less maintainable our code becomes in this example.
So we see we can play games with Perl data structures which reduces our overall memory usage, but to do so we have to sacrifice something else - programmer time - by producing less maintainable code. Which is more important in any particular example is a hard choice to make, but one you can make from a much more informed position with Devel::Size to help you.