Hello everyone!

I am a network field technician by day, and one of the sites I work on has a strange problem. The external Internet connection is about as stable as a tightrope walker on a tightrope in a dingy in a very violent storm. Quite a skilled one mind, because the connection doesn't actually go down, it just gets very slow every 3 seconds or so.

Enough of the back story. I'm trying to write a utility that will monitor the speed of a download over it's progress. I'm not interested in averages, which are comparatively simple to put together (just time a socket connection till it's finished, say) and can be done with tools like iperf. An average will just iron out the weirdness that I'm trying to detect.

At first I thought I could do something along these lines:

my $sock = new IO::Socket::INET->new(        # open up a socket to the
+ testing server
        PeerAddr => 'vm1.naxxtor.com',
        PeerPort => '80',
        Proto    => 'tcp',
        Type      => SOCK_STREAM
) or die("Can't open socket $!");
print "Sending request\n";
print $sock $request;                        # send the request for th
+e test file
print "Starting transfer\n";
{
    local $| = 1;
    my $chunk_start = 0;                     # initialise some variabl
+es
    my $chunk_time = 0;
    local $/ = \10240;                        # read in chunks 10240 b
+ytes long
    $chunk_start = [gettimeofday];            # set initial timer
    while (<$sock>) {                         # loop through each chun
+k    
        $chunk_time = tv_interval($chunk_start); # calculate the time 
+taken to download this chunk.
        if ($chunk_time != 0) { push(@speeds,$chunk_time); }    
        $chunk_start = [gettimeofday];                            
        
    }
}
[download]

Unfortunately this doesn't work - as sockets are BUFFERED, so if it recieves the data quicker than perl can get through the loop then you end up with a zero time for that chunk. It just goes plain screwy.

I have a horrible feeling I'm going to have to write this in C/C++ :(

What would probably work is to put a timestamp into a list for each chunk of 1024 bytes recieved, then process it later on. I can then do some windowing to get a reasonable estimation of speed at a particular point, and throw it out to a file which I can gnuplot.

I'm not sure. Any advice will be very well recieved :-)