Performance Monitoring Tools versus ..........

[KapilRaj]

Thanks to all who have replied. I was not in for 2 days hence missed this thread to have a look.

I have the following queries though my impression is changed towards the legasy monitoring tools.

1. According to one thread "Legacy monitoring tools are not updated to get the correct information" ? WHY ?. Does it mean that HP does not perform any more developement on legacy monitoring tools ?.
2. The amount of time taken to get the performance data is less compared to displaying it in a readable format. In that case Should I go for another HP Server to run the server agent (!?) of such monitoring tools ?.
3. I have noticed that the legacy tools does not produce correct o/p when the system has saturated. Does these monitoring tools help in monitoring at that time. Do they consume less CPU cycles and memory ?.

If you can still find some time for me ... please help me out.

Kap

[PVR]

Perfview with the help of measureware agents gives more accurate data compare to Glance plus. these agents collects the data store them in its datafiles, thereby offloads the burden from monitoring tools.(/var/opt/perf/datafiles)

HP is definitely working for improving their monitoring tools.. not their legacy tools.Finally money matters..

[KapilRaj]

I need more advice on this my dear friends . By the way am i asking tooooo much?.

Kaps

[Bill Hassell]

Legacy tools:

- These are industry standard tools and they are maintained in their standard format to remain compatible. Think of the confusion of tar or cpio all had new options and the tape format was changed. Same with top and iostat, etc. They work the way they worked (with some minor improvements below the surface) many years ago. They can't be fixed to provide the accurate and timely data needed without a total rewrite -- which is what Glance/Measureware are. Similar difficulties exist for Solaris which is why Glance/Measureware have been ported to that platform.

Measurement overhead:

No measurement tool can ever be non-invasive unless it passively tracks all the memory activities (ie, electronic instruments that probe the running computer). Putting the tool on another machine, even if you could transport the raw data to the remote system will be a huge performance penalty. It's not that the tools are slow but that things change in the kernel so fast that the tools must make assumptions and inferences. By the time Glance has put two characters on the screen, the CPU may have switched programs 10 times, completed a dozen I/O's and handled a meg or two of LAn data. The tool must remain as detached as possible to not impact the running kernel, something that legacy tools cannot do, thus the concept of midaemon and shared memory.

Leagcy tool accuracy

On a really busy system, legacy tools have no better chance in getting their neeeded CPU cycles than any other program, so they begin to bog down and accuracy (really, meaningfulness) will degrade. They use system calls like pstat and have to wait in line with all other programs that want their calls answered. With midaemon, many of the measurements are automatically collected within the kernel and midaemon just copies the data to shared memory along with important time references to keep the measurements straight. This is way beyond anything that legacy tools can do.

[Bill Hassell]

One other note: ALWAYS stay up to date on patches. And selective patching, ie, getting the latest 'top' patch but ignoring some kernel patches may still cause top to report inaccurate (sometimes negative) numbers. This is a case where "if it ain't broke, don't fix it" does not work because the fixes affect other tools. This also goes for Glance/Measureware. They should always be kept up to date on patches, right along with the kernel.

[Dwyane Everts_1]

Kaps,

I believe PVR hit the nail on the head...at some point HP has to look at making a profit, so the legecy tools aren't really updated since they are free.

1. According to one thread "Legacy monitoring tools are not updated to get the correct information" ? WHY ?. Does it mean that HP does not perform any more developement on legacy monitoring tools ?.

I think PVR answered this one.

2. The amount of time taken to get the performance data is less compared to displaying it in a readable format. In that case Should I go for another HP Server to run the server agent (!?) of such monitoring tools ?.

I have Glance Plus on a less utilized server collecting data from all my other servers. And Measureware agents are configured to send traps to Network Node Manager. The collection of data by these tools isn't very intense, but displaying the data will cause some overhead issue on some servers.

3. I have noticed that the legacy tools does not produce correct o/p when the system has saturated. Does these monitoring tools help in monitoring at that time. Do they consume less CPU cycles and memory ?.

The incorrect readings from your legecy tools during saturation is due in part because the tools are using the same resources that are saturated. Measureware detects an issue, and notifies you of the condition...not much overhead there. GlancePlus will see the same issues because it also uses the same resources.

Hope that answers some questions for you...

D

[KapilRaj]

Excellent replies!! as I expected form ITRC. Thanks to all.

The last request ( I am never satisfied (!?)) from me.

Can anybody refer a quick look document on Glance and Measure ware?.

You may suggest a configuration guide or some notes which you may have. I want to learn more about implementing this product quickly. (Trust me I am not appearing for an interview !).

Once again thanks to all the expert views.

Kaps

[Vijaya Kumar_3]

A great thing about Glance is, you can do your own scripting to collect any performance information.

I worked on a project to collect glance output and draw graphs (we dont have Measureware) and publish it in the web. Here are the useful forum topics:

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=243838

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=245388

Measureware is a another great tool for historic graphs.

Get all of your glance documents here:

http://www.docs.hp.com/hpux/netsys/index.html#HP%20OpenView%20GlancePlus%20and%20GlancePlus%20Pak


-Vijay

[keith persons]

Kapil,

There really aren't any more manuals available for Glance - years ago the effort was made to convert to on-line help files. They really are quite good in Glance. To try them out you can install the trial version of the product, you'll have 60 days to experiment. There are 2 types of help available, one for the keystrokes to get to different screens (if using the character version) and then the online help for the screens and metrics.

Glance is pretty straight forward. In brief, just install and then start - it will continue to gather metrics until you exit Glance.

Measureware is a little more involved and it is now part of the Performance Agent. You can examine some relevant manuals at http://docs.hp.com/hpux/netsys/index.html#HP%20OpenView%20Performance%20Agent%20for%20Unix
this should provide some additional insight into the Measureware functionality including a dictionary of all metrics that can be collected for all the supported vendors.

There is also another difference in the operations between the legacy commands and HP perf tools - if the monitored system is experiencing performance constraints, the collector process - midaemon - is designed to disable itself to prevent further degrading throughput. The legacy tools will try to run regardless.

Also, the HP perf products are also designed to consume (last I heard) no more than 10% os system resources, again, to keep from promoting performance problems.

Lastly, it's not usually suggested to rely exclusively on one product for perf metrics. This is to aid in insuring accurate data. You'll never get a one-to-one correlation since HP perf tools use logrithmic averages versus snapshots for the other commands. If severe discrepencies arise between the two it could be a trigger for more investigation.

Hope this helps.

Keith