[Orca-users] working with mixed-column data sources...

[David Michaels]

Attila Mezei-Horvati wrote:
> Thanks for the reply. I like the article and plan on
> checking the others, too. 
>
> My problem however is a little different. I am trying
> to monitor the performance of a web application on the
> server. It is called only from time to time, so I do
> have complete metrics (timestamp, startup time,
> communication with the network, operations, total
> duration) but I don't have it layed out based on a
> specific interval. Sometimes I have an entry at 5
> seconds, sometimes I have it after two three hours. I
> can set the interval to 5 seconds but then what should
> I use for the non-existing timestamps? I cannot put 0
> since that would alter my statistics, make it look
> like everything was alright (bringing down the total
> avg, too).
>   

It depends on what you're measuring.  Something like "cumulative number 
of system errors" is an example of a measure that only updates 
sporatically.  In that case, every 5 minutes you'd just record the 
current number of system errors.

In other cases, like website bandwidth, if your website receives very 
little traffic, you may have zero bandwidth for 3 hours, then 2 minutes 
of high bandwidth as someone visits your site, then 3 hours of no 
bandwidth.  Your "avg" statistics would probably have far less value to 
you than "burst" statistics.  On my webserver, the Orca graphs for such 
circumstances consist of nothing but thin-line spikes.  The latest rev 
of Orca implements RRD in a way that better maintains these mini-spikes; 
though they get flattened from averaging as the graph's period increases 
(daily -> weekly -> yearly), the relative representation is still 
useful.  As an example, I have a web serer that typically only gets a 
few hits in any 5 minute period, with most 5 minute periods having zero 
hits.  But even on the yearly graph, I can see this activity represented 
(as 200m instead of 2-3 hits, but I can still compare one month's 
activity to the next).

In the case of your web application, if you're trying to monitor its 
activity, you'll still see something show up on the Orca graphs, even if 
most of your timestamps have "0" in the column for that data.  If you're 
trying to gather statistics on it, that's another story--couldn't your 
statistical analysis simply ignore zero-value entries, and thus report 
only on "running periods" or such?  Or else, if Orca wasn't part of the 
equation at all, how would you normally account for the periods of 
inactivity?

--Dragon