[Orca-dev] [PATCH] Updates to include additional Solaris performancemetrics for graphing

[Blair Zajac]

Sean O'Neill wrote:
> 
> >Also, please include the log message with any diffs.  It makes it
> >easier to have the entire patch and log message in one email,
> >rather than finding the old email with the log message.
> 
> Not sure what you mean here.  What "log message" ?  You mean a log entry
> from svn ?  Assuming so, I thought log entries were only done during a
> commit action which I can't do - maybe time for me to read the SVN doco
> again ? :)

The log message you sent with the original email in November, portions
shown here:

Additions to Orca to collect and graph the following metrics for SE 3.2.1
(Solaris 8 and below) and SE 3.3 (Solaris 9):

pgrec - page reclaims (include pageout)
pgfrec - page reclaims from free list
.....
Changes made to:

* orcallator/orcallator.cfg.in:  additions to graphs metrics listed above

* orcallator/orcallator.se: additions to collect metrics listed above

* orcallator/start_orcallator.sh.in: updates to correctly identify the
version of the SE tool kit in use to properly include the correct version
of orca_p_vmstat_class.se.
.....


> >In lib/Makefile.in and ./orcallator/start_orcallator.sh.in, how
> >about we put the 3.2.1 and 3.3 directories in a SE directory,
> >so it looks like
> >
> >lib/SE/3.2.1
> >lib/SE/3.3
> >
> >and that it installs this way also.

OK.

> 
> k
> 
> >Regarding ./lib/Orca/Constants.pm, I'd like to keep this change out of
> >this patch.  How do the output PNG files look with this change?
> 
> http://chivas.oneill.dhs.org/orca/o_gauge_usr_pct,o_gauge_sys_pct,o_gauge_wio_pct,o_gauge_idle_pct-hourly.html

The plots look great.  Did you change the plot width on these or
did Orca make them wider automatically?

> 
> The data elements in the graph are still obvious enough that you can
> differentiate between the data elements for whatever interval you have Orca
> setup for.   Note the data in the graph example is 1 minute interval data -
> not 5 minutes.  I don't have any graph examples show this new data using 5
> minute intervals - sorry.
> 
> >One of the things that these numbers are designed for is that the number
> >of data points in a plot is less than the number of pixels in a plot,
> >otherwise RRD needs to resample the data, and this change from 1.5
> >days to 3 days bumps the number of data points from 432 to 864, which
> >is greater than the default width of 500.
> 
> Three days ??? The change I made to Constants.pm makes the Hourly graph
> chart 3 hours worth of data instead of the usual 1.5 hours - not 3 days -
> unless there is something I missed when changing this - which apparently I
> did.  You've lost me on the math you are using for the number data elements
> vs pixels.  Following the comments in Constants.pm:
> 

Oops, sorry, got the plots confused there.

Yes, no problem changing the number of hours in this plot as long
as Orca makes the width wide enough for all the data points.

> (1.5*60*60)/300 = 18 data points
> 
> My changes double this to 36 data points but I'm not sure how this relates
> to what you are referring to above.
> 
> How does this math work out when the collection interval is changed to 60
> seconds ?

I just compare the number of points being plotted with the number of pixels
in the output plot.  If they don't match well, then resampling occurs.

Here's the text from Constants.pm:

# For each plot type, the first value in the array reference is the
# number of 300 second intervals are used in a plot.  The second value
# is the number of seconds graphed in the plots.  Be careful to not
# increase the time interval so much that the number of data points to
# plot are greater than the number of pixels available for the image,
# otherwise there will be a 30% slowdown due to a reduction
# calculation to resample the data to the lower resolution for the
# plot.  For example, with 40 days of 2 hour data, there are 480 data
# points.  For no slowdown to occur, the image should be at least 481
# pixels wide.

> Actually, if this does something weird like forcing Orca to
> "resample" all the time, can you write something up and put this into the
> Orca FAQ to explain the "cost" of making the interval shorter - assuming
> this is a cost that is ?

Is the text above good enough?

Best,
Blair

-- 
Blair Zajac <blair at orcaware.com>
Plots of your system's performance - http://www.orcaware.com/orca/