Hi everyone,

We are currently trying to implement PMF 5.2.1 (environment: windows 2003 server/oracle 8 – WebFOCUS 7.6.10) but we have serious problems with response times.
We don’t understand why, because we are currently testing PMF with ONLY (for the moment) 1 scorecard, 3 measures and 15 dimensions (of course each measure doesn’t use all dims).

We bought 300 licenses (so 300 users can “log in” potentially at same time) but the good news is we estimated that maximum 70 users will potentially use it AT THE SAME TIME.

Based on that, we have some questions for PMF customers:
-How many dimensions and measures are you using in your datamart? Which size (in number of rows)?
-How many users are using PMF? And how many of them should use it concurrently?

We are wondering if other customers are encountering same response time issues within PMF as we do. As an example, when 5 users are accessing the same dashboard (4 gadgets) simultaneously, it generally takes more than 2 minutes to get a server response but it’s too long for the dashboard to wait (it then displays a warning popup message, when not crashing).

By our side, the “measures” table in the database has 2 300 000 rows (with 3 measures), Our goal is to deal with 42 measures…

Any help would be appreciated.
Thanks.
Frenz

2:00 minutes. That's 1:50 too long! PMF dashboards should display in seconds. In the lab, and at customers where we've given some consultancy to improve performance, PMF delivers complex dashboards in <10 seconds.

If such time not your result, please have a look first at the PMF Installation and Configuration Guide, and refer to the sections on performance. There are good diagnostics and practice notes on how to get PMF to perform well at your site. And read on, please.

PMF is a complex and large Web application. It cannot perform fast on its own without some time spent on each component to optimize the environment to the expected performance standard you want to set. Left to its own devices, without any attention paid to performance, any large Web application will founder. Ask Google, or Yahoo, or any company responsible for creating a fast and responsive app. If your servers are running slow, or your RDBMS, or your network, then each will slow down the performance. We'd assume there are multiple choke points slowing down various components of the app in your environment, which explains the far below acceptable response time.

We'd be happy to show you PMF running in our lab, where we get dashboards back fast, just to prove it can be done. We also recommend you call in to IB/ATS, and schedule some time to go over your environment with the performance experts there.

DIAGNOSING APP PERFORMANCE
In general, diagnosing app performance issues is a practice and is iterative. Depending on the performance and response time you expect, you might need to look at every possible choke point and optimize it. Performance is never automatic. It must be designed into the app (which it is with PMF) and then enforced in the environment.

Initial question to ask is, what is the expected performance of the app in seconds? (Besides saying "fast," what does "fast" mean?)

Once you know the expected speed, you should go over the entire setup of the app environment, so the possible performance choke points can be explored.

A) DB Server. DB Version, processor cores, available memory.

B) Reporting Server. WF Version, processor cores, available memory, number pre-started agents, max agents. Is tracing turned on?

C) Web server box: Processor cores, available memory.
* Web server: Version, configuration. Running HTTPS or HTTP? Is HTTP compression turned on?
* App server (usually Tomcat): Configured heap size. Is tracing on?

D) Browser: Type and Version. Is the site configured to the browser as “Local Intranet?” Is caching turned on? If Flex is being tested, is Flex caching turned on?

E) Network: Speed of pipes between each choke point. Any special items to take note of? For example, are you logging on through a security bridge or switch that might be slowing down traffic? How is the person doing the test accessing the app to do the test? T1 line? 10baseT, 100baseT, gigabit, wifi B/G/N?

Once you know the answers to each of these, there are many steps you can take to optimize each. There are caching mechanisms, memory allocation, task priority allocation (depending on the platform), undburdening various choke points if they are carrying too much non-PMF traffic, etc.

thanks - hope this helps

While you're gathering that information, you can do some additional diagnostics that might help you determine if the issue is due to slow performance from the RDBMS.

Note that PMF asks the RDBMS to perform the summarization of your metrics. That's because to transfer possibly millions of rows of metric data over to WebFOCUS to perform the summarization is a guaranteed slow performer - sure, once WebFOCUS has the records in memory, summarization is almost instantaneous, but there's the "small" issue of moving millions of records over the wire... would not work well.

So the SQL issued from PMF is optimized to request summaries from the RDBMS on the dimensional intersections used for the metrics.

You can check the timings for your RDBMS query by running a PMF report or View, then in your browser doing a View Source and looking at the traces (below the final

</html>

tag).

Look for the following:

-* +++++++++++++++++++++++++++++++++++++++++++ Run time for Step(s) Extract Normal Measure Data : nn.nnn seconds

How many seconds is showing for nn.nnn? If it's more than 3 seconds, the RDBMS definitely needs to be optimized.

Please refer to the following FocalPoint articles on optimizing RDBMS performance with PMF:

#1 RDBMS Performance - Part 1
#2 RDBMS Performance - Part 2
#3 RDBMS Performance - Part 3
#4 RDBMS Performance - Part 4 -- Specific techniques recommended

Hope this helps.

Hi,


In the PMF Installation and Configuration Guide -> Optimizing PMF Performance -> WebFOCUS Application Server Configuration (page 118).


The guide talk about create a separate service for PMF. We do that, but how can we force the agent to use this service ?

Because, the service is determined :
-by the client profile with the IBI_domaine variable
or
-by forcing the IBIC_server variable

Isn'it ?

Our question is :
How to force the &IBIC_server= PMFservice to every asynchronous fex start from the mainstreet application ?

Thanks,

Carlos
BCEE Luxembourg

All PMF content comes through the Performance Management Framework Domainat runtime.

In the MR Administrator, There is a setting for each Domain to select a default Service.

Select the Service you created for te Performance Management Framework Domain. That's all there is to it.

thanks

Hi Bob,

That what we do, but in our case only the fex that are in PMF basedir passed through the PMF service. All the gadgets of every dashboard fallen into the default service.

Is this the expected behavior ? Or is only specific in our installation ?

Thanks,


Carlos
BCEE Luxembourg

Every online request from PMF should go through with the domain flags set to on. You might want to check your site.wfs to see if you're doing something to defeat that.

I'm not the world's greatest WebFOCUS configuration person but that seems like a good first thing to check.

There should be a section in site.wfs that checks for the PMF domain. All focexecs run with that set from the domain.

thanks