Any ideas on how to figure this out?

Yes. Use top or tasklist to monitor both the perl executable and the java executable during the command execution; for cpu, IO & memory usage.

In theory, the Perl instance should show negligible activity in all three categories. If that is not the case, post your findings and someone here may be able to shed some light on the possibilities.

Apart from the timing issues, the logic you're using to build the command line is strange. I would have done it something like this:

my $pathstr = join( '', $outputfiles, $invoice_data_ref{actual_invoice
+_id} );
my @cmdargs = ( 'java', '-jar', '/var/ww/helper_code/svg2pdf.jar',
                "$pathstr/master*.svg", "$pathstr/master.pdf" );

print join( ' ', 'command_str:', @cmdargs ), "\n";  # optional - just 
+to check

$report_failure = system( @cmdargs );
[download]

my @svg_names = ( qw/master01 master02 .../; # or whatever those names
+ are
# … create those files ...

my $pathstr = join( '', $outputfiles, $invoice_data_ref{actual_invoice
+_id} );
my @cmdargs = qw/java -jar /var/ww/helper_code/svg2pdf.jar/;
push @cmdargs, "$pathstr/$_.svg" for ( @svg_names );
push @cmdargs, "$pathstr/master.pdf";
[download]

I expect that after you generate all those "fairly complicated SVGs", you close those file handles (i.e. make sure all the output has been flushed) before doing the system call. If those file handles are still open, perl will flush them for you when you do the system call, which might add a bit of delay.

Your description suggests that the system call is only being done once; if you were doing a lot of similar system calls in some sort of loop, then one possible cause of delay would be the extra overhead of invoking and cleaning up after a subshell on each iteration; one way to get around that (on a *n*x system) would be:

open( SH, '|-', '/bin/sh' ) or die "Can't open a shell: $!";
for my $iterator ( @whatever_set ) {
    print SH "command line…\n"
}
close SH;
[download]

#!/usr/bin/perl
use strict;
use warnings;
system( '...' );
exit;
[download]

If I hard code the command it still takes 13 seconds. And I'm sure the delay is coming from just this call because I'm timing it by calling time() just before and just after the call to system.

EDIT: Kindly disregard this post as it has become moot in view of the reply to the preceding post that in fact answers my question.

This might be a dumb question, but how are you determining the execution times? If you are using the obvious method of time command on the command line then you are including the startup time for perl, which may be significant depending upon hardware, OS, etc.

Java is notorious for it's slow startup. If you execute java hepler for some web-applications from user requests, it could be performance killer for your application (e.g. as I think, if you run 10 java instances in simultaneously)

So, if you have access to java-sources and know java a little bit, to rewrite your helper as network daemon, listening local interface and communicate with it via sockets. At least problem with "startup overhead", which you attribute to perl, will disappear.