Quarterly Chart on Performance

[Madanagopalan S]

I am preparing the Quarterly Performance Chart and
including CPU,MEMORY,DISK and Network. I have lot
of metrics logged thru adviser for three months for the
above.

Can anyone help me to make this chart more readable
and understandable for capacity planning? What are
all the important metrics, I need to look?? Pls help
me..
Thanks in Advance.

[Stefan Farrelly]

What is adviser ??

The important stats to present when doing weekly or monthly views is to show wether resources are maxing out at any point. If you simply average everything over such a long period the graphs will look fine and any maxing out will not be shown as it gets averaged out.

So, for example, if CPU runs at 100% for 10 mins during peak times during weekdays I would want any graph to show this. Same for disk and memory, if any resource reaches maximum for a period I want to see it. Averages over such a long period will not show this.
Then I could plan for upgrades/resolutions to fix it.

Showing the main metrics, Cpu, Memory, Disk (space and utilisation by disk) and Network are fine. Possibly also kernel tables to see if any are nearing full.

[Madanagopalan S]

Thanks for your mail.

I run glance in adviser mode and collected some metrics for cpu,disk,memory,network and system tables.

Now, I have the metrics for each minute. How can make it useful and efficient chart? Anybody done this?? Can anyone tell your experience on this.

[Stefan Farrelly]

The HP tool designed to present collected stats is called PerfView (now OV PerfView). This is a graphical tool which presents color graphs in a huge array of different options. This is what we use.
Doing it using adviser manually is going to be a long and difficult task. I recommend you buy Measureware and PerfView and let them do the work for you....

[Tommy Palo]

With MS Excel you can import ASCII-files and make nice charts, I do that myself for some statistics collected over time, not with Glance though.
You could probably quite easy format the stat. files in unix for that purpose.

[Madanagopalan S]

Stefan,
I cannot go for Measureware now, due to budget.
Measureware running and its logging statistics (using
export). Are u using this??

Tommy,
If you don't mind, can you give me a example
of your statistics, may be I can get some idea by
seeing.

Anyone having any chart for the performance statistics.

Thanks in Advance.

[Stefan Farrelly]

This is what the output from the perfstat command shows as running on our servers; Midaemon is the real-time collector for glance/gpm. Scopeux is the long-term stats collector which you use PerfView (pv) to graph.

**********************************************************
** perfstat for hal2 on Thu Nov 23 09:18:28 GMT 2000
**********************************************************

list of performance tool processes:
----------------------------------

Running midaemon (measurement interface daemon) pid 1374

MeasureWare scope status:
Running scopeux (MWA data collector) pid 1377

MeasureWare background daemon status:
Running ttd (Transaction Tracker daemon) pid 1339

MeasureWare server status:
Running alarmgen (alarm generator) pid 1798
Running agdbserver (alarm database server) pid 1797
Running perflbd (location broker) pid 1376

The following data sources have running repository servers:
PID DATA SOURCE
Running rep_server 1796 SYSDOWNT
Running rep_server 1795 LVOLUMES
Running rep_server 1764 SCOPE


We also run the DSI integration module which allows even more customised stats to be collected and graphed, and alarms for ITO to monitor.

If I had to do it manually without measureware and perfview I would let gpm run all day and at the end of it capture the Alarm History window which lists all bottlenecks for the day (network, cpu, disk and memory) and either print it or capture the window in Windows/NT (via an X emulator or use XV on hp-ux) and paste it into a word document and print it in color and email it to my manager. This should be easy and quick to do.

[Madanagopalan S]

I have attached my machine perfstat status. I compared both In my machine, all the processes are running.
How can I take the statistics. Can u help me??

I tried to run gpm for whole day for one week but I
didn't get any alarm history.
**********************************************************
** perfstat for host1 on Thu Nov 23 17:26:44 SST 2000
**********************************************************

list of performance tool processes:
----------------------------------

Running midaemon (measurement interface daemon) pid 1162

Running gpm (glance motif interface) pid 18845

Running glance (glance character-mode interface) pid 4414

MeasureWare scope status:
Running scopeux (MWA data collector) pid 1194

MeasureWare background daemon status:
Running ttd (Transaction Tracker daemon) pid 1106

MeasureWare server status:
Running alarmgen (alarm generator) pid 2079
Running agdbserver (alarm database server) pid 2078
Running perflbd (location broker) pid 1157

The following data sources have running repository servers:
PID DATA SOURCE
Running rep_server 1982 SCOPE


ITO Agent Status:

WARNING: opcctla is not running
WARNING: opcmsga is not running
WARNING: opcacta is not running
WARNING: opcle is not running
WARNING: opcmona is not running
WARNING: opcmsgi is not running
WARNING: opcuisrv is not running


******** (end of perfstat output: note above warning) ********

[Stefan Farrelly · Accepted solution]

\You have the scopeux collector running which means you are collecting data, good. The logfiles are in /var/opt/perf/datafiles, they are called logxxx, you should see them and they should be quite big in size. The best way to produce stats from these logfiles is to use PerfView. But you can extract some data manually using the extract command (see man page for it and manual on docs.hp.com - theres a lot to read about it before you start using it), but to graph it you need to do it yourself using excel or some other tool.

If you ran gpm for a day or week and got no alarm history thats good! that means your HP server is performing really well, no bottlenecks at all. You should be happy and report to your manager that it is performing well with no problems at all.

[Madanagopalan S]

Stefan,
Can u do me one more help?? Can u send me
your perfview graph? So that I can get a idea,
I can use those metrics from the graph.

Thanks for your help.