[smokeping-users] Re: Another suggestion...

[Tom Throckmorton]

On Jan 28 21:51, Hal Burgiss wrote:
> As I see it there are three possible interesting statistics:
> 
>  routing latency
>  network stack latency
>  server response latency

Yep, but not only is defining the metrics for measuring these difficult, as evidenced by multiple competing
measurement efforts (e.g. iepm, ipma, ippm), but measuring all three simultaneously and accurately is tricky!

I think defining 'server response' is one of the hard ones, however I _have_ found it useful/practical sometimes
to also gauge the response of a particular service, which sometimes exposes (or exonerates) a network latency
issue.  As an example, i created a monitor that would fetch (and time) a given url across the network, and then
compared that to the result of fetching the same url locally (on the system that hosts the url).  What i found in
one case was that the response time of the remote fetch (across the network) took 2-3 times longer than the local
fetch, which showed nicely how much the network was contributing to the web server's responsiveness (or lack
thereof :-).  I had a different case though where the server response times were both high and matched tightly,
which turned out to be a database performance/latency problem.  The point here is, there can be a bunch of
mitigating factors when looking at server response times, so if you want to really see what the problem is if you
think your network is chewing, sometimes you have to dig a little deeper than ICMP....


-tt

-- 
Tom Throckmorton
Harvard Business School
Network Operations Center
(Avoid contact with skin)

--
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