Monitoring DMF tape drive use with PCP

File kmchart.png is a screen shot showing the monitoring of DMF tape
drive use, with CPU use in our two cpusets shown on the same time scale.

    Local background - CPUs 12-127 on "cherax" are reserved for user
    batch work, using torque and moab.  CPUs 0-11 are for everything
    else: system, interactive, cron & DMF.

Colours are:
	CPU	red		system
		white		interrupt
		blue		user
		yellow		wait I/O
		green		idle
	Drives	red		down
		blue		in use by DMF
		pale blue	idle, (dis)mounting or non-DMF

This lets you see interesting things like:

- the surge in wait-I/O time corresponding to writing to T10000B tapes

- the fact that our T9940B's are flat out

- multiple drives going down simultaneously (we were having troubles
  there)

- If you knew our cron and RUN_TASK schedule, you can glean other
  factoids from these charts.


To do something similar for yourselves:

- make sure that your DMF machine is running PCP, with the pmdadmf2 PMDA

- install kmchart on your desk/laptop system.  A comment from the
  kmchart maintainer:

    - The "kmchart" program was renamed "pmchart" (at SGI's request) a while
      back, and they are slowly phasing out the old with the new (free) one.
    - The "kmchart" package (rpm, deb, etc) was renamed "pcp-gui" a while
      back too, looking ahead to the pmgadgets and pmview replacements which
      we will want all in the one package.
    - The Windows version is available now - there's a single package for
      Windows (named "pcp-glider") which includes all of pcp, pcp-gui, and
      all of the infrastructure PCP needs to run on Windows (cc, cpp, perl,
      sh, awk, sed & other shell tools, etc).

  Look in your distribution's standard repositories, or in
  ftp://oss.sgi.com/projects/pcp/download/ .  Note that sourceforge
  only has a very old version.

- copy file idle_drives.kmc to directory ~/.pcp/kmchart, changing the
  hostname ("cherax") and drive names (starting with "cd" and "bd").
  Add/subtract lines to reflect number of CPUs on your system.

- run:

	kmchart -h cherax -c idle_drives.kmc -t 150

The resulting window unfortunately takes 24 hours to become fully
populated, so just leave it running.

Enjoy.