Upload
tamsyn-howard
View
224
Download
4
Tags:
Embed Size (px)
Citation preview
• What are the barriers to real-time
business intelligence (BI)?
• How can Microsoft SQL Server 2005
Analysis Services be used to make BI
more real-time?
What Will We Cover?
Difficulties of Real-Time BI
SSISSSIS
POS OLTP Cleanse andEnrich
DW
POS OLTP
UDMStaging
Cube
ValidationUDMProduction
Cube
Barriers to Real-Time BI
Barrier Solution
Historical view None
Coordination with business processes
across systems
None
Data quality management None
Integration of multiple data sources Heterogeneous Query Processing (HQP)
Data consistency Snapshot Isolation
More Barriers to Real-Time BI
Barrier Solution
Managing aggregates
Isolating the OLTP system from long-
running queries
Data Push & Proactive Caching
Knowing what has changed Notification services
Linking back to the source system Actions
Pushing Data into UDM
Data can be pushed directly into a Unified Dimensional Model
SQL Server 2005 Integration Services processing transforms
Includes fact and dimension tables
SSISSSIS
POS OLTP Cleanse andEnrich
POS OLTP
UDMProduction
Cube
Demo
Linking Integration Services (SSIS) and Analysis Services Directly
View an SSIS Package Run an SSIS Package Browse the Updated Cube
demonstration
Updating with Trickle Feeds
Trickle feeds can get data directly into the UDM
Integration Services updates the cube every few minutes
SSISSSIS
POS OLTP
POS OLTP
UDMProduction
Cube
Cleanse andEnrich
Building the Cube Directly
UDM can combine data from multiple sources
One of the underlying sources must be SQL Server
Not applicable for all scenarios
SSISSSIS
POS OLTP
POS OLTP
UDMProduction
Cube
Cleanse andEnrich
Continuously Changing Data
Problem Solution
How to handle updated data
Source data might be continually changing
How to ensure consistency during processing
Use Snapshot Isolation
Proactive Caching
• Policy-based managementHas source data changed?When to refresh?How to answer queries during refresh
• Proactive caching combinesOLAP query performanceReal-time data access as needed
• No more explicit “cube processing”
Using Policies to Refresh the Cache
UDMUDM
POS OLTP
POS OLTP
Policy-based refresh of the cache
UDMProduction
Cube
Demo
Using MOLAP and Reverting to ROLAP when Latency Exceeded
View Partition Settings Cause Latency Revert to ROLAP
demonstration
Proactive Caching Challenges
• EfficiencyHow to avoid overloading Analysis Services with frequent updates
How fast can the caches catch up?
• PerformanceHow to balance between latency and performance
• NotificationsIs the cache refreshed on change or periodically?How does AS know that the RDBMS has changed?
Policy Settings
Property Description
SilenceInterval After an update, for how long must there be a quiet time with no further updates before rebuild starts?
-1 (infinite) = no quiet time
SilenceOverrideInterval If no quiet time, start anyway after this time
-1 (infinite) = no override
ForceRebuildInterval How long after last cache was built should rebuild of a new cache always commence?
-1 (infinite) = no periodic rebuild
Latency How out-of-date can the cache be before reverting to ROLAP?
-1 (infinite) = never revert to ROLAP mode
Scaling Up
Problem Solution
How to handle large quantities of data
Re-creating the whole cache on every change is expensive
Use ROLAP
Use partitions
Use incremental cache updates to add data