There are times when planning and forecasting databases grow for apparently no reason at all. The static data (YTD actuals) that is loaded hasn’t changed and the users say they aren’t doing anything different.

If you load budgets or forecasts to Essbase, you probably do what I’m about to tell you. If you are a systems administrator and have never seen how finance does a budget or forecast, this might be an education.

The culprit? More...

There are several ways to export data from Essbase on a large scale. Pulling it via Excel (Smart View or the Essbase Add-In) is not the best way to get large amounts of data when the goal is to move the data somewhere else, so this option will not be covered.

Database Export

The easiest method is to export all the data from a database by exporting the database.  This can be done in EAS.  This method is easy to automate with Maxl, but has little flexibility with formatting and the only option is to export all the data.  It can be exported in column format so the data can easily be loaded into another data repository.  If the data needs to be queried, or manipulated, this is a good option.  More...

The format of the data that is loaded to Essbase is often an after-thought.  But, should it be?  When requesting the data file from a source system, it is more important than you may think to have it sorted to mirror your outline.    

Assume an outline has the following dimensions.

• Period [DENSE]
• Account [DENSE]
• Region [SPARSE]
• Category [SPARSE]
• Product [SPARSE]
• Organization [SPARSE]

The most efficient way to receive a data file would be to have it sorted by Organization, Product, Category, Region, and then Account.  Data files load faster when the columns that hold the sparse members are sorted in reverse order of the sparse dimensions that exist in the outline.

The reason the data loads faster is because it opens a block of data only one time.  If the data was sorted by the dense members first, then every block would have to be opened multiple times.  If the same sparse member combinations have 3,000 dense members with data, the block would be opened up to 3,000 times.  

There are some more important benefits of doing this, however.  When the block is opened multiple times, the database becomes far more fragmented than it needs to be.   Fragmentation causes calculations to be slower and retrieving data can also be impacted, which can lead to frustrated customers.

By not sorting the data when loaded, every time a data load occurs, any performance issues that may exist are exacerbated.  So, anytime possible, sort the data load files by the last sparse dimension in the outline, the second to last sparse dimension in the outline, and so on.  You may be presently surprised at the benefits.

Everybody knows the quickest way from point A to point B is a straight line.  Everybody assumes that the path is traveled only one time – not back and forth, over and over again.  I see a lot of Essbase calculations and business rules, from experienced and novice developers, that go from point A to point B taking a straight line.  But, the calculation travels that line multiple times and is terribly inefficient. 

Here is a simple example of a calculation.  Assume the Account dimension is dense, and the following members are all members in the Account dimension.  We will also assume there is a reason to store these values rather than making them dynamic calc member formulas.  Most of these are embedded in a FIX statement so the calculation only executes on the appropriate blocks.  To minimize confusion, this will not be added to the example.

Average Balance = (Beginning Balance +Ending Balance)  / 2;
Average Headcount = (Beginning Headcount + Ending Headcount) / 2;
Salaries = Average Headcount * Average Salaries;
Taxes = Gross Income * Tax Rate;

One of the staples of writing an effective calculation is to minimize the number of times a single block is opened, updated, and closed.  Think of a block as a spreadsheet, with accounts in the rows, and the periods in the columns.  If 100 spreadsheets had to be updated, the most efficient way to update them would be to open one, update the four accounts above, then save and close the spreadsheet (rather than opening/editing/closing each spreadsheet 4 different times for each account).   

I will preface by stating the following can respond differently in different version.  The 11.1.x admin guide specifically states the following is not accurate.  Due to the inconcistencies I have experienced, I always play it safe and assume the following regardless of the version.

You might be surprised to know that the example above passes through every block four times.  First, it will pass through all the blocks and calculate Average Balance.  It will then go back and pass through the same blocks again, calculating Average Headcount.   This will occur two more times for Salaries and Taxes.  This is, theoretically, almost 4 times slower than passing through the blocks once.

The solution is very simple.  Simply place parenthesis around the calculations.

(
Average Balance = (Beginning Balance +Ending Balance)  / 2;
Average Headcount = (Beginning Headcount + Ending Headcount) / 2;
Salaries = Average Headcount * Average Salaries;
Taxes = Gross Income * Tax Rate;
)

This will force all four accounts to be calculated at the same time.  The block will be opened, all four accounts will be calculated and the block will be saved. 

If you are new to this concept, you probably have done this without even knowing you were doing it.  When an IF statement is written, what follows the anchor?  An open parenthesis.  And, the ENDIF is followed by a close parenthesis.  There is your block!

"East"
(IF(@ISMBR("East"))
    "East" = "East" * 1.1;
ENDIF) 

I have seen this very simple change drastically improve calculations.  Go back to a calculation that can use blocks and test it.  I bet you will be very pleased with the improvement.

There is, what appears to be, a bug in Hyperion Planning that causes business rules that take longer than 5 minutes to re-launch.  The following, published by Oracle, explains the root issue of this problem.  It is not a bug, but a setting in the host web server that causes the request post multiple times.  This explaination from Oracle clearly states that this is ONLY an issue when accessing Hyperion Planning through Hyperion Workspace.  I have seen the same response while accessing Hyperion Planning directly.  Regardless of your entry point, it is a good proctice account for either entry method and should be applied.

This applies to Hyperion Planning, Version: 9.3.1.0.00 to 11.1.1.3.00 and is applicable to all operating systems.

Symptoms

When accessing Planning, Business Rules that normally take more than 5 minutes to complete
run for an unlimited period of time.  By viewing the running Essbase sessions in the EAS console, you can see that the Business Rules "Calculate" sessions are being re-launched every 5 minutes, so that a new instance of the Rule is launched before the first can complete.

This issue only affects Business Rules that normally take more than 5 minutes to complete.

This issue does not affect Business Rules launched directly from Planning (accessing Planning directly on its own URL, bypassing the Workspace).  This issue does not affect Business Rules launched from the EAS console.  This issue only affects systems using Weblogic as a web application server.

Cause

This issue is caused by a default timeout setting of 5 minutes (300 seconds) in the Weblogic HTTP Server Plugin.  This plugin is a set of configuration files in which Weblogic defines how it will interact with the HTTP Server through which Workspace is accessed.  More information on Weblogic Plugins is available here:  http://download.oracle.com/docs/cd/E13222_01/wls/docs92/pdf/plugins.pdf

Solution

Hyperion System 9 and Oracle EPM 11.1.1.x support the use of either Microsoft Internet Information Services (IIS) or Apache as an HTTP server. The steps to increase the timeout depend on which you are using.  The new timeout value should be set to a value larger than the time the longest-running Business Rule takes to execute. The examples below use a setting of 30 minutes (1800 seconds).

Apache HTTP Server

Step 1

Edit %HYPERION_HOME%\common\httpServers\Apache\2.0.52\conf\HYSL-WebLogic.conf

Step 2

Add (or edit, if already present) the following parameters to the two sections for Planning, and also to the two sections for Financial Reporting and Workspace, as the 5 minute timeout issue can cause problems in all three products.Each section begins with an XML tag.

WLIOTimeoutSecs 1800
HungServerRecoverSecs 1800

<LocationMatch /HyperionPlanning>
<LocationMatch /HyperionPlanning/*>

Add the new "WLIOTimeoutSecs 1800" and "HungServerRecoverSecs 1800" properties as new lines within the tags.  If you are using a version of Weblogic prior to 9.x you need to add the second line "HungServerRecoverSecs 1800" in addition to the "WLIOTimeoutSecs 1800" parameter. This second parameter is not necessary for Weblogic 9.x and later (though it will do no harm).

PathTrim /
KeepAliveEnabled ON
KeepAliveSecs 20
WLIOTimeoutSecs 1800
HungServerRecoverSecs 1800

Internet Information Services (IIS)

Step 1

There are several copies of the iisproxy.ini file. Oracle recommends you modify the files for Planning, Financial Reporting and Workspace, as the 5 minute timeout issue can cause problems in all three products.

Paths (note that "hr" below stands for Financial Reporting):

%HYPERION_HOME%\deployments\WebLogic9\VirtualHost\hr
%HYPERION_HOME%\deployments\WebLogic9\VirtualHost\HyperionPlanning
%HYPERION_HOME%\deployments\WebLogic9\VirtualHost\workspace

Step 2

For each copy of iisproxy.ini, add the following lines at the end of each file.  If you are using a version of Weblogic prior to 9.x you need to add the second line "HungServerRecoverSecs=1800" in addition to the "WLIOTimeoutSecs=1800" parameter. This second parameter is not necessary for Weblogic 9.x and later (though it will do no harm).

WLIOTimeoutSecs=1800
HungServerRecoverSecs=1800

Step 3

Restart IIS from the IIS Manager and restart the Workspace web application service

Oracle HTTP Server is used

Step 1

Modify the file mod_wl_ohs.conf file under the directory, $EPM_ORACLE_INSTANCE\httpConfig\ohs\config\OHS\ohs_component with the following content:

<LocationMatch ^/HyperionPlanning/>
SetHandler weblogic-handler
WeblogicCluster PlaningServer:8300
WLIOTimeoutSecs -1
WLSocketTimeoutSecs 600
</LocationMatch>

Step 2

Restart the Oracle HTTP server and the Workspace web application services after the modifications are complete.

Changes to an Essbase outline cause changes to the Essbase index and data files, regardless of the method (Essbase Administration Services, Hyperion Planning database refreshes, or from a script).

Changes that require restructuring the database are time-consuming (unless data is discarded before restructuring).  Understanding the types of restructures and what causes them can help database owners more effectively manage the impacts to users.

TYPES OF RESTRUCTURES

Essbase initiates an implicit restructure after an outline is changed, whether done with the outline editor, through an automated build, or some other fashion like a Hyperion Planning database refresh.  The type of restructure that is performed depends on the type of changes made to the outline.

DENSE RESTRUCTURE:  If a member of a dense dimension is moved, deleted, or added, Essbase restructures the blocks in the data files and creates new data files. When Essbase restructures the data blocks, it regenerates the index automatically so that index entries point to the new data blocks. Empty blocks are not removed. Essbase marks all restructured blocks as dirty, so after a dense restructure you must recalculate the database. Dense restructuring, the most time-consuming of the restructures, can take a long time to complete for large databases.

SPARSE RESTRUCTURE:  If a member of a sparse dimension is moved, deleted, or added, Essbase restructures the index and creates new index files. Restructuring the index is relatively fast; the time required depends on the index size.

Sparse restructures are typically fast, but depend on the size of the index file(s).  Sparse restructures are faster than dense restructures.

OUTLINE ONLY:  If a change affects only the database outline, Essbase does not restructure the index or data files. Member name changes, creation of aliases, and dynamic calculation formula changes are examples of changes that affect only the database outline.

Outline restructures are very quick and typically take seconds.

Explicit restructures occur when a user requests a restructure to occur.  This can be done in Essbase Administration Services or via Maxl (and EssCmd for those of you who still use it) and forces a full restructure (see dense restructure above).  It is worth noting that this also removes empty blocks.

CALCULATING IMPLICATIONS AFTER RESTRUCTURES

When a restructure occurs, every block that is impacted is tagged as dirty.  If Intelligent Calculations are used in the environment, they don’t provide any value when a dense restructure occurs as all blocks will be calculated.  When member names or formulas are changed, the block is not tagged as dirty.

WHAT DICTATES THE RESTRUCTURE TYPE

The following outline changes will force a dense restructure, which is the most time- consuming restructure.

DENSE AND SPARSE

• Defining a regular dense dimension member as dynamic calc
• Defining a sparse dimension regular member as dynamic calc or dynamic calc and store
• Defining a dense dimension dynamic calc member as regular member
• Adding, deleting, or moving dense dimension dynamic calc and store members
• Changing dense-sparse properties [Calc Required]
• Changing a label only property [Calc Required]
• Changing a shared member property [Calc Required]
• Changing the order of dimensions [Calc Required]

DENSE (DATA FILES)

• Deleting members from a dense dimension  [Calc Required]
• Adding members to a dense dimension
• Defining a dense dynamic calc member as dynamic calc and store member

SPARSE (INDEX)

• Adding members to a sparse dimension
• Moving members (excluding shared members) in a sparse dimension
• Defining a dense dynamic calc member as dynamic calc and store
• Adding, deleting, or moving a sparse dimension dynamic calc member
• Adding, deleting, or moving a sparse dimension dynamic calc and store member
• Adding, deleting, or moving a dense dimension dynamic calc member
• Changing the order of two sparse dimensions

NO RESTRUCTURE OCCURS

• Deleting members of a sparse dimension [Calc Required]
• Deleting members of an attribute dimension
• Deleting shared members from a sparse or dense dimension [Calc Required]
• Adding members to an attribute dimension
• Adding shared members to a sparse or dense dimension
• Moving a member in an attribute dimension
• Renaming a member
• Changing a member formula [Calc Required]
• Defining a sparse dynamic calc member as dynamic calc and store member
• Defining a dense or sparse dynamic calc and store member as dynamic calc
• Defining a regular dense dimension member as dynamic calc and store
• Defining a sparse dimension dynamic calc and store member or dynamic calc member as regular member
• Defining a dense dimension dynamic calc and store member as regular member
• Changing properties other than dense-sparse, label, or shared [Calc Required]
• Changing the order of an attribute dimension
• Creating, deleting, clearing, renaming, or coping an alias table
• Importing an alias table
• Setting a member alias
• Changing the case-sensitive setting
• Naming a level or generation
• Creating, changing, or deleting a UDA

WHAT DOES THIS MEAN

Understanding this can help users and administrators manage applications to better meet the needs of all those involved.  When designing an application, knowledge of this topic can be instrumental in the success of the application.  Here are some things to keep in mind.

• When updating an outline or refreshing a planning application, it may be faster to export level 0 (or input level) data, clear the data, perform the update, and reload/aggregate the export when  changes cause a dense restructure.
• For dimensions that are updated frequently, it may be beneficial to define those dimensions as sparse.  Changes to sparse dimensions typically require only restructures to the index file(s), which are much faster.
• If frequent changes are required, enabling incremental restructuring may make sense.  Using this defers dense restructures.  The Essbase restructure happens on a block by block basis, and occurs the first time the data block is used.  The cost is that calculations will cause restructures for all the blocks included and the calculation performance will degrade.
• Setting the isolation level to committed access may increase memory and time requirements for database restructure.  Consider setting the isolation level to uncommitted access before a database restructure.
• If multiple people have access to change the outline, outline logging may be useful.  This can be turned on by adding OUTLINECHANGELOG = TRUE in the essbase.cfg.
• Monitoring progress of a restructure is possible when access to the server is granted.  Both sparse and dense restructures create temporary files that mirror the index and data files.  Data exists in the .pag files while indexes are stored in .ind files.  As the restructure occurs, there are equivalent files for each (pan for data files and inn for index files).  In total, the restructure should decrease the size of the ind and pag files, but the pan and inn files can be used for a general idea of the percent of completion.