Data distribution by frequency

[Greg Fortune]

My cable access through the dorms at EWU provides me with between 5-10 k/sec
with up to 40% packet loss to my first hop (main router) on bad days and
generally above 5%.  At about 3:00AM, I can get around 20k transfers and I
got really lucky one night and pulled 44k through Windows and 77k through my
Linux box.  Needless to say, that sucks.  I've finally got irritated enough
to write some code on an idea I had last year regarding the same issue.  I'm
going to generated statistics on our local network (time intervals we are
down, ping times, packet loss) and make that information available to the
university throughout the year.  Hopefully, that will force the university
to actually *do* something about our access.

Right now, the only things I haven't written are the timing loop and an
interface to pull the info back out of the database.  The timing loop is
needed to keep my network activity to between 20-30 packets every 10
minutes.  I think that should show sufficient results, but only if the key
machines (ie, the blasted DNS server and main router) are pinged more often
than selected outside hosts (ie yahoo.com, linux.com, etc).  Thus, I need to
use some kind of frequency (or maybe probability density function?) to
determine how often a target should be pinged.

I guess that I could just hard code everything in, but I've got this silly
idea that I might use a user customizable timing loop again at some point
<g>.  Who knows, but at least it sounds good.

I'm not familiar with the term "probability density function" at all.  If
that seems like a simpler way for me to accomplish what I need, please
explain or point me to resources.

And you're right about the amount of code :)  Even the way I came up with is
fairly straightforward.

Thanks,

Greg Fortune


<snip>