[smokeping-users] Any downsides to running less probes more frequently?

[G.W. Haywood]

Hi there,

On Fri, 25 Jul 2008, Jim Long wrote:

> IANA statistician, nor am I a smokeping guru, but...

Sometimes you can get carried away with satistics.  After all, it was
said "there are lies, damned lies, and statistics"... ;)  I'd be very
reluctant to translate one lost packet PER MINUTE out of five (or four
out of twenty) as 20% packet loss, when the packets I'm measuring are
56-byte UDP pings and there are probably tens of thousands of 1.5kByte
TCP packets (mostly SMTP and HTTP in my case) passing PER SECOND over
the same routes at the same time.  I think that the effective TCP loss
is zero, as all the nightly backups that I do (they range from about a
hundred megabytes to some tens of gigabytes after compression by bzip2
or gzip) make it in tact to the backup servers.  I don't have a good
handle for the numbers of TCP retransmissions, nor of how much rsync
has to do on top of that; if I thought there might be a problem that's
where I'd look next.  None of this will be constant even over a short
timescale: a quick Google for "tcp congestion algorithm" (without the
quotes) gave me about 338,000 hits.

In the case of your network, the best thing to do is probably what you
find works best for you.  Only you can know how your network is being
used, and what is appropriate for one network might not be relevant to
another.  For example I'm using a lot of different probes (UDP pings,
DNS pings, TCP pings for SSH and HTTP) from a few different machines
to a lot of other machines; my graphs are all on logarithmic scales;
and now I've got it mostly buttoned down I don't look at the graphs as
often as I used to.  I've started to remove the debug logging.  That's
because I've thought about what the networks are doing, taken a view
on what I need from the tools, and now I think I've got it, I'm happy.

I wouldn't presume to say that what I've done would be the 'best'
way of doing it, even supposing I had any idea what that might be,
but It Works For Me.

Thanks again, Tobi, BTW. :)

--

73,
Ged.