Performance monitoring

[Anil C. Sedha]

1. What is the best way to monitor increasing system resource utilization - memory, CPU, disk usage, disk hits for short term analysis.

2. Same question but for long term predictment & analysis. Also i need to know which processes cause this increase. I would like to plot them later on to a graph and analyze.

Pls exclude perfview, SAR in your replies. info on openview would be appreciated. If you could provide some of the great knowledge that each of you possess would be really helpful.

Thanks
Anil (All replies will be duly appreciated and honoured with points)

[Zafar A. Mohammed_1]

Checkout the MeasureWare(MWA) product.


Thanks
Zafar

[Helen French]

The best answer for your question would be GlancePlus. Reasons can be:
1) It's an Openview product.
2)It can give most relaible and accurate information about resource usage.
3)This will be useful for short term as well as long term analysis.
4) Just go through these feautures for a better idea:
http://managementsoftware.hp.com/products/gplus/index.html

[Ramkumar Devanathan]

Hi Anil-jee,

You could try the Unix OS SPI and its measureware/glance integrations to give you a complete monitoring over your entire node's hardware (disk, cpu, memory, swap, filesystem (hw!), network interfaces).

It'd create alarmdef's as you configure in the templates. Of course, it assumes that measureware/glance is deployed on the nodes.

So, install mwa/glance on your node and then, deploy the osspi template groups for mwa/glance onto the node - follow the unix os spi admin ref for the exact details.

let me know if you require any more details. the unix os spi is present in the OV applications cd.

Unix OS SPI also provides for monitoring with coda and without any of the performance products.

- ramd.

[Anil C. Sedha]

Thanks for the messages upto now.

Let me be more specific. I am aware of perfview(mwa), glance.

This requirement is more of a high level data gathering for predicting future requirements on our systems, analyzing and reducing constraints due to apps.

Hope this will help you to give me more detailed info. Where are the pharaoh's hiding today ( ;-) )

Regards,
Anil

[Tim Sanko]

Just because You have chosen not to pick SAR may I suggest that I might trend SAR logs and show increase in DISK IO, CPU, etc. If you keep the info in spreadsheets you can easily import it and graph it for Management consumption.

if the memory line inclines at a 15 degree angle then Upper management can understand it.

If DiskIO hits 100% on the graphs everyone understands it...

[PIYUSH D. PATEL]

For the long term perdictment and analysis, you would need to export the MWA data and store it for further review. Since you are using Perfview, you would be aware of this. You can also use another product : HP Performance Manager which is a Web based software and anybody can access the Performance graphs through Java/web based interface. You can even design your own metrics and drill down to the process level and save those metrics for future use.

Thanks,
Piyush

[Martin Johnson]

At appears you are looking for a Capacity Planning product. Check out BMC's Best/1.


HTH
Marty

[Mark Laser]

top: Lets you see what is going on in the system

prospect: Uses kernel instrumentation to read/time event records kept by the kernel

tusc: prints out system calls and is able to trace signals

glance: gives a subset of what prospect gives

Mark Laser

[Bill Hassell]

Glance gives you a quick snapshot while MeasureWare collects long term data. Tools like sar use legacy (read: old) interfaces into ther kernel which lack the detail that you need. The MeasureWare logs have hundreds of metrics that are not otherwise available through freeware and legacy Unix commands.

Once you have MeasureWare installed, just use the extract command to pullout data ready for import to spreadsheet programs. PerfView is somewhat awkward to use when compared to popular spreadsheet programs like Excel. You can assign groups of programs (ie, an instance of Oracle or an accounting package) to be treated as a special category to track usage based on that category, all standard in MeasureWare.

[Decio Miname · Accepted solution]

This is another tool that worths a look:
http://www.automatos.com/
It has been provided by HP as a licensed service after the Compaq thing.
The highlight of this tool is its long-term capacity planning analysis. It creates up-to-date reports that go way long beyond load/memory/disk/etc increase tendencies - it even tells you about hardware models and configuration that would supply the future needs when the thresholds are reached. I am not expert on this area, but I've never seen such details on any other tool.
For short-term analysis and performance troubleshooting, Automatos is quite basic. For that, it has no comparison with Glance, for example.

[Suhas_2]

Hi Anil,
I have been using BMC's Best1 software for past 2 years and my experience is fantastic. The separation of utilization of CPU is very clear. We can separate out utilization of each process. Data can be easily archived. The graphs can be easily generated and published on the web.
For capacity planning we use the "predict" module of Best1.
If you need online information about the utilization on the system or the utilization by the process then there is another module called PATROL. It can be used for system monitoring and can send alerts that can be caught by HP-Openview or IBM-Tivoli.
Do check-out BMC's site for further information.
(www.bmc.com)
Hope this helps...
Regards..
Suhas

[Paddy_1]

Measureware should be a good tool for your needs.Coming to the analysis part this is what I would suggest.Import measureware data into a Data Warehouse(you can Use SAS if you are really keen of doing stat analysis).Once you got data in your ware house then you can start slicing and dicing using some kind of OLAP analysis which should give you a good idea of the solution.