Well, i'd written a similar utility to scan around 10,000 devices for ping,
I used Threads with semaphore. Since it's windows, be careful about concurrent threads.
The whole operation took around 20 mins. But i had a limitation of memory consumption
as well. Here are the modules i used:

use threads qw(yield);
use Thread::Semaphore;
use Net::Ping;
use Socket;
[download]