Performance Tuning HFM in the Field
description
Transcript of Performance Tuning HFM in the Field
HUGmn 2010 Jim Heflin
Chris Barbieri
Application Design: the Foundation of Performance
Hyperion Financial Management
Metadata design as it impacts performance
Data volume and content measurement
Rules performance measurement
Reading the HFM logs
Designing HFM’s 12 Dimensions
Application Profile1. Year2. Period3. View
System4. Value dimension,
includes currencies
User controlled5. Entity6. Account7. ICP8. Scenario
User defined9. Custom 110. Custom 211. Custom 312. Custom 4
Application Profile
YearNo inherent impact on performanceCannot be changed after the application is builtImpacts the number of tables that can be
created in the databasePeriod
The base periods comprise the column structure of every table, whether you use them or not.
For this reason, avoid weekly or yearly profiles unless it is key to your entire application’s design
ViewNo impact, but only YTD is stored and Periodic,
QTD are on-the-fly derivations
What’s a Subcube?
Parent subcube, stored in DCN tablesCurrency subcubes, stored in DCE tables
Metadata Volumes (Americas)
Data Design
DensityContent
Specifically: zerosTiny numbersInvalid Records
Loaded Data
What percent of the loaded data is a zero value?No hard rule, but <5% may be reasonableNo zeros are best, watch ZeroView settings on the
scenariosWatch out for tiny values, resulting from allocationsHow much does the data expand from Sub Calculate?
Am I generating zeros, or tiny numbers?
Data Density Using FreeLRU
Survey of data density using FreeLRU method
Number of applications reviewed: 32 Average Min Max Median ABC Customer
NumCubesInRAM 2,672 72 10,206 1,345 577
NumDataRecordsInRAM 1,502,788 247,900 5,627,748 1,170,908 1,107,614
NumRecordsInLargestCube 86,415 2,508 593,924 53,089 31,446
Average records per cube 6,309 24 91,418 1,352 2,288
Average metadata efficiency: average cube/densest cube
7.3% 0.3% 39.7% 3.4% 7.3%
HFM 11.1.1: the magic of 64 bit!
Measure and Analyze RulesHow much time
do I spend in each rule?
Let’s focus on the “top
10”
Rewrite the rule for optimal
performance
Do some months take longer than
others?
Is it because they have
more data?
Establish a Baseline
Effect of cachingData cache on database server AND on HFM application
serverCaches may be empty during first runPerformance is significantly better when data reads comes
from memory cache rather than disk This is why cache management is so important
Run the same process 3 times in a row and use the average
“Performance begins with perception. Establish this and a baseline before applying science.”
Chris BarbieriSr. Product Issues Manager
Hyperion SolutionsMarch 5, 2006
“Rules” of Thumb
Most application between 0.25 and 2.0 seconds per 0.25 and 2.0 seconds per entity, per periodentity, per periodConsolidate all with data for entire hierarchy, full yearDivide by total number of entities (descendents of
selected parent), divided by 12 periodsMost applications are closer to 0.25 seconds
Rules Impact RatioBlank rules file, Consolidation Rules = N for baselineDivide consolidation time with rules by time withoutUsually 2-5 times2-5 times
Data Density <> Calc Time
The Black Art of Reading HFM Event Logs
Where does HFM store its event information?
Maintaining the logsHow can I view this?OK, what does it actually tell me?
Understanding HFM Logs
Messages Messages are informational –start/stop consol, log in, log
out etc. Some messages are purposely out of time order (consol
starts get printed at completion of consol
Warnings Often due to subcube size issues HFM Subcube Troubleshooting Guide / Memory
Management in HFM documents
Errors Access rights Syntax Issues
Where are the HFM events stored?
Text file containing XML, named HsvEventLog.log
Pre-HFM 9.2.0.2 or 9.3.0 ..\Hyperion Solutions\Hyperion Financial Management\Server Working Folder\
Starting with 9.3.1 Oracle moved all product logs to a common parent folder HYPERION_HOME\Logs\FinancialManagement or HYPERION_HOME\Logs\HFM
How can I view this?
Administration Module Web: Administrators only
HFM Error Log Viewer utility Free standing executable Bundled with HFM under \Consultant
Utilities
Web System Messages
Available to administrators
Launch the Utility
Launch HFMErrorLogViewer.exe
System Message panel
Details panel
Details
Web suppresses richer details shown in utility
Find “Registry”
Each server’s registry settings are written during an application start-up.
Most but not all registry entries are writtenWe’ll cover the actual entries in another
presentation
System Memory at Inception
Page File Size Increased in 9.2.0.3, 9.3.1 to 130 and 260 MBStill exists in 64 bit HFM 11.1.1, but likely unused
Paging
Watch “PageOutOps > 0” indicating page file usage
Consolidation start and finish
Summary indicates start time
Details have finish time
Is written when it completes
Extracting Log Entries
HFM writes to both the event log and the database
You can extract the database entries to a text file, which is preferable to the event logs
Can also truncate the entries using this utility
And split large files (anything > 30 MB is too large)
Ranzal Performance Lab Team
Chris Barbieri Established HFM
performance tuning techniques and statistics widely used today
4+ years as Sr. Product Issues Manager at Hyperion
Member of HFM launch team in 2001, certified in HFM and Enterprise
MBA, Babson College B.S. Finance &
Accounting, Boston College
Co-founded in 2007
Kurt Schletter Over 20 years in IT Hyperion Support Manager
at United Technologies, serving 3,600+ HFM users
5+ years Hyperion product infrastructure services
MBA, Rensselaer Polytechnic Institute
B.S. Management with Computer Applications, Worcester Polytechnic Institute
Visit www.Ranzal.com/News.htm for a listing of complete webinars
Chris BarbieriChris [email protected]@ranzal.com
Needham, MANeedham, MAUSAUSA
+1.617.480.6173+1.617.480.6173www.ranzal.comwww.ranzal.com