thetechnologyevaluator’s...
TRANSCRIPT
The Technology Evaluator’s Cheat Sheets
Business Intelligence & Analy:cs
WWW.SISENSE.COM
Summary
• So1ware Stacks – Full Stacks (DB + ETL Tools + Front-‐End So1ware) – Back-‐End Stacks (DB and/or ETL Tools Only) – Front-‐End Stacks (Front-‐End So1ware Only)
• Technologies – Data Warehouse Class (“Big Scale”) – Data Mart Class (“Small Scale”)
WWW.SISENSE.COM
So1ware Stacks
DW
ETL
BACK-‐END STACK ETL Features
Data Warehouse Features Data Mart Features
FRONT-‐END STACK Data VisualizaLon Features
Data Analysis & Discovery Features
ETL
Query/Import
FULL-‐STACK
WWW.SISENSE.COM
The Full Stack. When? • Centralized data management and storage
– To deliver a single version of criLcal data – To make data easier for non-‐techies to access, query and share – To simplify on-‐going or ad-‐hoc data management tasks
• ETL Func:onality Is Needed – MulLple data sources, or mulLple tables where views are too complex/slow – The volume of data is expected to cause slow performance – Data needs to be restructured before being delivered to users – Data is dirty (entry errors, value mismatches) – Required metrics are in different tables or sources
• To protect the opera:onal systems from rogue queries • To access non-‐queryable data sources
WWW.SISENSE.COM
End-‐Users (Business)
Data Warehouse + Data Marts Data Extracts (No DW)
DW
OLAP Cubes, or In-‐Memory Marts
End-‐Users (Business)
Data Sources
ETL / Mash-‐up
In-‐Memory Marts Excel/CSV
IT Department
Data Warehouse ETL / Mash-‐up
Data Sources
IT Department
Front-‐end Tools Front-‐end Tools
Full Stack: TradiLonal Architectures
WWW.SISENSE.COM
Data Warehouse: Pros & Cons DW + Data Marts Data Extracts (No DW)
Approach SoluLon-‐oriented Project-‐specific
Data Quality & Accuracy Higher Lower
Scalability Higher Lower
Single Version of the Truth Yes No
IniLal Investment Higher Lower
Level of Detail Summarized Granular
Owner IT IT or Business (opLonal)
ImplementaLon Time Longer Shorter
Technical Complexity Higher Lower
Advantage / Disadvantage
WWW.SISENSE.COM
Technologies In The Space
WWW.SISENSE.COM
Backend Technologies • Data Mart-‐Class, we call it “Small Scale” – Online AnalyLcal Processing (OLAP) – In-‐Memory Databases (IMDB)
• Data Warehouse-‐Class, we call it “Big Scale” – Database So1ware Appliances – Database Computer Appliances – Distributed Databases
WWW.SISENSE.COM
Small Scale. When?
• When there is only a single data source, which means the data doesn’t need to be consolidated (ETL) prior to being delivered for business analyLcs
• When there aren’t many different abributes and metrics to cross-‐reference (the Data Mart doesn’t need to have many fields)
• For a one-‐Lme project (e.g. one dashboard), with no added requirements, new data sources or other changes expected in the future
WWW.SISENSE.COM
Big Scale. When?
Big Scale Small Scale
Max. Data Mart Size Terabyte -‐ Petabytes Gigabytes
Max. Number of Fields (1 mart) PracLcally Unlimited Limited
Max. Number of Records (1 table) Billions Millions
• For a single centralized data store to serve mulLple users and mulLple business scenarios (single version of the truth)
• When data volumes are large, are rapidly growing or may unpredictably spike
WWW.SISENSE.COM
Data Mart-‐Class Technologies (“Small Scale”)
WWW.SISENSE.COM
In-‐Memory Databases (IMDB)
• Achieves fast performance by loading the enLre data mart into RAM, thus avoiding slow disk-‐reads (“I/O Boblenecks”)
• Categorized as “Small Scale” because the size of data mart is effecLvely limited by the size of RAM, placing in the Gigabyte scale category
• In some IMDB technologies, RAM consumpLon is also drasLcally affected by concurrent use.
WWW.SISENSE.COM
Online AnalyLcal Processing (OLAP) • Achieves fast performance by pre-‐calculaLng metrics (field
aggregaLons) for all sets and subsets of unique values in all dimensions (fields) ‘over-‐night’. This avoids performing these slow operaLons in real-‐Lme during the work-‐day.
• Categorized as “Small Scale” because storing the results of these pre-‐calculaLons (“The Cube”) takes exponenLally more storage resources than the actual raw data does, limiLng the actual size of raw data that can make up a cube to GB scale.
• The query engines behind most OLAP technologies are based on RDBMS technology, whose own scale and performance limitaLons OLAP cannot overcome (e.g. joining, grouping).
WWW.SISENSE.COM
Data Warehouse-‐Class Technologies (“Big Scale”)
WWW.SISENSE.COM
So1ware Appliances
A so1ware appliance is a soUware applica:on that might be combined with just enough operaLng system (JeOS) for it to run op:mally on industry standard hardware (typically a server) or in a virtual machine.
WWW.SISENSE.COM
Computer Appliances
A computer appliance is generally a separate and discrete hardware device with integrated so1ware (firmware), specifically designed to provide a specific compuLng resource. Computer appliances are generally not designed to allow the customers to change the so1ware, or to flexibly reconfigure the hardware.
WWW.SISENSE.COM
Distributed Databases
A distributed database may be stored in mulLple computers, located in the same physical locaLon; or may be dispersed over a network of interconnected computers. A distributed database system consists of loosely-‐coupled sites that share no physical components (such as disk, RAM and CPU)
WWW.SISENSE.COM
Big Scale Technologies, Compared SoUware Appliance
Computer Appliance
Distributed Databases
Hardware Class Commodity Proprietary Commodity
Best Architecture 1 Server 1 Server N Servers
Capacity Terabytes Terabytes Petabytes
Hardware Cost 4-‐5 figures 6-‐7 figures 5-‐6 figures
WWW.SISENSE.COM
Full-‐Stack Vendors ETL SoUware
Appliance Hardware Appliance OLAP IMDB In-‐Chip
SiSense Elas:Cube Manager
Elas:Cube (Columnar) -‐ -‐ -‐ Elas:Cube
(Columnar)
Microso1 SSIS SQL Server (RDBMS) -‐ Analysis Services PowerPivot -‐
Oracle Oracle ETL Features
Oracle DB (RDBMS) ExaData Hyperion ExalyLcs -‐
IBM InfoSphere DataStage
DB2 (RDBMS) Netezza Cognos Cognos -‐
SAP NetWeaver BW ETL -‐ HANA
Columnar In-‐Mem BW
Business Objects HANA
Columnar In-‐Mem -‐
Microstr. MSTR ETL Features -‐ -‐ Microstrategy
In-‐Mem OLAP Microstrategy In-‐Mem OLAP -‐
QlikView QlikView Expressor -‐ -‐ -‐ QlikView
AssociaLve In-‐Mem -‐
WWW.SISENSE.COM
Thank You! Visit us at www.sisense.com