Veritas-bu

[Veritas-bu] Graphing NetBackup Statistics with RRD TOOL (wha t to graph?)

2002-02-06 07:12:34
Subject: [Veritas-bu] Graphing NetBackup Statistics with RRD TOOL (wha t to graph?)
From: plb AT iotk DOT com (Peter L. Buschman)
Date: Wed, 06 Feb 2002 13:12:34 +0100
Tim:

>I am also looking at creating graphs for netbackup and other services.

Great!  Maybe we can pool our efforts :)

>I am looking at using RRD Tool with Cacti as the front end. You said
>that there were other graphing tools which you could use, what are these
>tools?

I haven't tried the Cacti front end.  Is it nice?   I'm actually 
masochistic enough to use RRD TOOL's
obtuse graphing syntax directly (now that was a learning curve.)

Since I collect all my statistics using Perl and NBUX, I should be able to 
use any of the graphing
modules from CPAN.  RRD TOOL is really only good for time-series data like 
counters where you
are interested in the change in a variable over time.

         http://search.cpan.org/Catalog/Graphics/

The only downside to that is that many of these don't work on both Unix and 
NT.  However, for the
moment, neither does RRDtool.  The statistics piece, however, which uses 
NBUX, should work
reasonably cross-platform.

>Can you explain how you generate the graphs, I presume you have a script
>which runs every 5 minutes and then invokes rrd tool to give it the new
>data?

Almost.  I have two scripts, one which creates and logs data into the RRDs, 
and another which creates
graphs as .gif files  using RRD TOOL's graph function.

I use a 5-minute step interval, but don't require my logging script to run 
every 5 minutes.  Instead, the
script is a little more intelligent and will check the time the RRD was 
last updated and loop over every
5 minute interval from the last update to the present, plugging the stats 
in at each step.  This way,
resource-intensive commands like bpdbjobs only need to be run once to 
update multiple steps.  I use
a heart-beat of 48 hours so as long as the update script is run once every 
48 hours, the RRDs can be
kept up to date.  In practice, once an hour is generally good, but it can 
be more or less frequent depending
on your needs and the performance impact.

--PLB


>Timothy
>
> > -----Original Message-----
> > From: veritas-bu-admin AT mailman.eng.auburn DOT edu
> > [mailto:veritas-bu-admin AT mailman.eng.auburn DOT edu] On Behalf Of
> > Peter L. Buschman
> > Sent: 06 February 2002 11:31
> > To: SIBLEY, Ken R. - ACCOREL; 'NetBackup List'
> > Subject: RE: [Veritas-bu] Graphing NetBackup Statistics with
> > RRD TOOL (wha t to graph?)
> >
> >
> >
> > Ken:
> >
> > Drive utilization is an easy one.   I might do that next.*
> >
> > Throughput statistics per drive are easy on Solaris (iostat
> > -xnM) but much
> > tougher on other systems since
> > NetBackup doesn't track this data.  Does anybody know if
> > there are ways to
> > get this data on Windows?
> >
> > Throughput per job is an easy calculation to make, but you
> > end up with an
> > average over the entire run of
> > the job, not the peaks and valleys over time.  RRD TOOL has
> > ways of working
> > around this, but the database
> > would have to be populated at every step interval to track
> > the growth of
> > the kbytes counter in the job record.
> > (This would probably necessitate running bpdbjobs too
> > frequently.)  Right
> > now, I can update on 5 minute intervals
> > going back 48 hours, but only having to run bpdbjobs once.
> > My logging
> > script is very efficient and not tied to
> > the time interval it graphs.
> >
> > Average throughput per job is a great one to add to a gannt
> > chart that
> > shows the time period for which the
> > job ran.  I like this thought a lot (a colored line for each
> > jobid and the
> > average throughput as a subscript.)
> > RRD TOOL isn't the right graphing tool for this one, but
> > there are other
> > graphing modules out there.*
> >
> > The wheels are really turning now..... there are some interesting
> > directions we could go in terms of "backup visualization."
> >
> > --PLB
> >
> >
> > At 03:38 PM 02/05/2002 -0600, SIBLEY, Ken R. - ACCOREL wrote:
> > >I would be very interested to see (in order of priority):
> > >Drive Utilization over time (very big)
> > >Throughput Statistics (per drive and per client/class)
> > >Job Statistics
> > >Number of available/up/down drives
> > >Client/Class active over time
> > >
> > >I think these would be great.
> > >
> > >Ken
> > >
> > >Ken Sibley
> > >Sr. Unix Administrator
> > >Accor Economy Lodging
> > >ksibley AT accorel DOT com
> > >469-737-3370
> > >
> > > > -----Original Message-----
> > > > From: Peter L. Buschman [mailto:plb AT iotk DOT com]
> > > > Sent: Tuesday, February 05, 2002 1:34 PM
> > > > To: Veritas-bu AT mailman.eng.auburn DOT edu
> > > > Subject: [Veritas-bu] Graphing NetBackup Statistics with RRD
> > > > TOOL (what
> > > > to graph?)
> > > >
> > > >
> > > >
> > > > All:
> > > >
> > > > I've recently been experimenting with using RRD TOOL for
> > > > storing NetBackup
> > > > statistics.
> > > > I finally got my first pair of jobs stats loggers and
> > > > graphers working
> > > > tonight.  You can see
> > > > the results here (although it isn't very impressive given my
> > > > small system
> > > > at home.)
> > > >
> > > >          http://www.iotk.com/nbux/rrd/jobstats.gif
> > > >
> > > > I think this type of time-series data tracking could be a valuable
> > > > operational resource for
> > > > backup administrators.  In that regard, I'd like to poll the
> > > > list and ask
> > > > what statistics would
> > > > you like to see tracked over time?
> > > >
> > > > The ones that come to my mind are:
> > > >
> > > >          + Job Statistics (extending my first example to
> > > > include per-error
> > > > tracking.)
> > > >          + Throughput Statistics (how much data am I backing
> > > > up over time?)
> > > >          + Media Statistics (how many full / expired / frozen
> > > > / suspended,
> > > > etc, tapes in my media database.)
> > > >          + Size of catalogs.
> > > >          + Number of available / up / down drives.
> > > >          + Drive utilization over time.
> > > >
> > > > ...and much, much more.  If you can quantify a variable at a
> > > > point in time,
> > > > it can be stored in an RRD and graphed.
> > > >
> > > > If you've ever seen NetBackup Advanced Reporter, you have an
> > > > idea of the
> > > > data that can be collected and
> > > > processed.  NBAR is an expensive option, however, and not all that
> > > > customizable.  I'd like to add to the tools
> > > > available at no cost to the user community.
> > > >
> > > > I'd be very interested to hear everyone's thoughts on this.
> > > > How useful
> > > > would you find this kind of extension
> > > > to the base NetBackup functionality?
> > > >
> > > > Best regards,
> > > >
> > > > --PLB
> > > >
> > > >
> > > >
> > > > _______________________________________________
> > > > Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
> > > > http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
> > > >
> > >_______________________________________________
> > >Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
> > >http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
> >
> > _______________________________________________
> > Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
> > http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
> >
> >