[smokeping-users] "how to read it" Part II

Wed Aug 27 21:19:03 MEST 2003

And this is my answer to a few of his questions

tobi

-- 
 ______    __   _
/_  __/_  / /  (_) Oetiker @ ISG.EE, ETZ J97, ETH, CH-8092 Zurich
 / // _ \/ _ \/ /  System Manager, Time Lord, Coder, Designer, Coach
/_/ \.__/_.__/_/   http://people.ee.ethz.ch/~oetiker   +41(0)1-632-5286

---------- Forwarded message ----------
From: Tobias Oetiker <oetiker at ee.ethz.ch>
To: Arturo Martin G. <Arturo.Martin at sync.com.ve>
Date: Thu, 12 Jun 2003 23:25:56 +0200 (MEST)
Subject: Re: Smokeping : "how to read it"

Today Arturo Martin G. wrote:

>
>          Tobi : thanks so much for th clarification !!!!
>
>          I still have some questions ... see in [ ]
>
> At 11:13 p.m. 11/06/2003 +0200, you wrote:
> >Today Arturo Martin G. wrote:
> >
> > >
> > >          Hello Tobi :
> > >
> > >
> > >          First and foremost THANKS for the RRD/MRTG creation : probably the
> > > most popular piece of great software ever written for monitoring purpouses.
> > >
> > >          One of the biggest benefits of the original MRTG "traditional"
> > > graphs is its EASE to understand : it doesn´t take a Phd to make sense of
> > > most of the graphs. I've trying to understand the meaning of Smokeping and
> > > I just give up : don´t know if my speculations are correct.
> > >
> > >          Don't you have a "how to read an smokeping graph / potential
> > > applications for smokeping" kind of document ? I review everything online
> > > and (in your web page) I found nothing explanatory in this area.
> >
> >Hi Arturo,
> >
> >How about this, I will answer your questinons and you write a few
> >paragraphs so that others can profit too ?
> >
> >
> >Smokeping is a latency measurement tool. It sends test packets out
> >onto the net and measures the ammount of time they need to travel
> >from one place to the other and back.
> >
> >For every round of measurement smokeping sends several packets. It
> >then sorts the different round trip times and selects the middle
> >one, this means when there are 10 times, time number 5 is selected
> >and drawn (median). The other times are drawn as successively
> >lighter shades of gray in the background (smoke).
>
> [ Regarding the shades of gray : the fact that, when it happens, it goes
> from the darkest black to white has some implication with the actual
> distribution of the samples for the period (like more samples in a
> particular value means darker ?)

each sample gets a shade of grey, so that you can distinguish them
... if you can see a lot of smoke, this means that the samples are
fahr apart (20ms 30ms 80ms 130ms ....) not a lot of smoke means the
sample are close together ( 30ms 30ms 31ms 33ms 33ms 34ms )

> ? Or are just a way of saying : "hey, the
> round trip times for this particular time period are distributed SOMEHOW
> from the median up to X ms more / less".

not just somhow, it shows how the distribution is, because each
sample gets its own shade, so

30ms 30ms 30ms 60ms looks different from 30ms 40ms 50ms 60ms

> By the way, is very strange, if
> the MEDIAN is the line drawn, not find more often (at least in the
> examples) that the remaining samples for the period (20 every 300 sec, in
> the examples) are always ABOVE the median, and only rarely BELOW, but not
> frequently ABOVE and BELOW the median. By definition the median of Y
> samples should imply aprox. Y-1 "bigger" and Y-1 smaller, right ? (My
> speculation - waiting for your confirmation - is that when you do the
> measures you have found that the "smaller than median samples" tend to be
> very close to the median itself, which is a possible explanation of the
> reason for this observation)

exactly, this is the cool thing about the median, it is a good way
to get rid of 'bad' data ...  so if 15 samples come in at 30ms and
the remaining 5 are scattered between 40ms and 120ms this means
that the median will be 30ms and no smoke will be below the median
...

>
> Have you consider to include the MODE (not the average & median) for the
> samples ? ]

Excuse my ignorance, how do you calculate the MODE ? can't there be
several modes ? In the 'latency comunity' it seems that the median
is the accepted methode to look at the data ...

> >Sometimes a test packet sent out but never returns. This is called
> >packet-loss. The color of the median line changes according to the
> >number of packets lost.
> >
> >All this information together gives an indication of network
> >health. packet loss for example is something which should not
> >happen at all, it can mean that a device in the middle of the link
> >is overloaded or some routing configuration is wrong.
> >
> >Regarding round trip time, if this fluktuates heavily (lots of
> >smoke) this also indicates that the network is overloaded.
> >
> >Smokeping also has special probes to not just test roundtrip time
> >of the packet, but rather have it perform some task at the remot
> >end, eg download a webpage ... this will give a combined 'picture'
> >of webserver availablility and network health.
> >
> >smokeping can also look at this data all on its own, and when it
> >sees a predefined pattern, it can send out email ... a pattern may
> >be packet loss going from < 1% to over 20% or something like this
> >...
>
> [ What is the explanation for the background color of the graph and the
> leyend "Link up: XXX days / hours " ?? I assume it means : "for this period
> of time the IP address you are testing (or web page or whatever you are
> testing) has been DOWN for less than (eg) 1day bt more than 12 hours ? How
> do you calculate the "link up: XXX" value for the graph ? ]

well this is for links that change their IP ... (some adsl
providers do that to their customers (to me at the time smokeping
was written)) with this color you can see how long you were able to
retain the same IP, this only works in connection with the DYNAMIC
option in the config file
>
>
> >any more questions ?
>
>
>          By the way, I just love the photos on your web page of your trip
> around the river Inn : is that from your hometown to Austria ? Great photos
> man !!!! (must be the hell of a bike ride !!!)

My hometown (Olten) is in Switzerland. That bike trip started in
switzerland, went then to austria and germany  ... it was a great
time ... and a good photo oportunity on top of it ... hacking away
all day long I don't get to take that many pictures ...

cheers
tobi

>
>          Saludos
>
>          Arturo
>
>
>

-- 
 ______    __   _
/_  __/_  / /  (_) Oetiker @ ISG.EE, ETZ J97, ETH, CH-8092 Zurich
 / // _ \/ _ \/ /  System Manager, Time Lord, Coder, Designer, Coach
/_/ \.__/_.__/_/   http://people.ee.ethz.ch/~oetiker   +41(0)1-632-5286

--
Unsubscribe mailto:smokeping-users-request at list.ee.ethz.ch?subject=unsubscribe
Help        mailto:smokeping-users-request at list.ee.ethz.ch?subject=help
Archive     http://www.ee.ethz.ch/~slist/smokeping-users
WebAdmin    http://www.ee.ethz.ch/~slist/lsg2.cgi