Heads you’re in; tails you’re out: How RCTs have evolved in DWP
HEADS OR TAILS? - The SAS Ottawa Platform User Society · HEADS OR TAILS MODE : NON-DISTRIBUTED OR...
Transcript of HEADS OR TAILS? - The SAS Ottawa Platform User Society · HEADS OR TAILS MODE : NON-DISTRIBUTED OR...
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS?
DECISIONS AROUND SAS® VISUAL ANALYTICS
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS DECISIONS AROUND SAS® VISUAL ANALYTICS
There are many important decisions that you face
when implementing SAS® Visual Analytics. This
short session describes some of them, and provides
background information that will make those
decisions more than just a flip of a coin.
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS DECISIONS AROUND SAS® VISUAL ANALYTICS
Option 1 Option 2
Mode Non-Distributed Distributed
Version VA VAAR
Metadata Server Shared Separate
Hadoop SAS Provided Commercial
VA Location Asynchronous Co-Located
LASR servers One Many
Working with Data Data Builder External
Data Formats ABT Star Schema
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS MODE : NON-DISTRIBUTED OR DISTRIBUTED
Non-Distributed
Single Server
Lowest cost of entry
Ceiling to growth
Slower reload of data to memory
SAS Visual Analytics Environment
Metadata Server
Visual Analytics
Middle-Tier
Workspace Server
Visual Analytics has its own metadata server for managing the SAS VA environment
2 x # Core x64 Processors## GB RAM * / Linux x64
* Minimum Recommended hardware
SAS LASR Analytic
Server
SAS Visual Analytics Environment
Node n
Node 4
Node 3
Node 0
Metadata Server
Visual Analytics
Middle-Tier
Workspace Server
2 x # Core x64 Processors## GB RAM * / Linux x64
Node 2
SAS Hadoop/HDFS
Data Node
2 x # Core x64 Processors## GB RAM */ Linux x64
SAS High-Performance Analytics Environment
Worker Node
SAS 9.4
Visual Analytics has its own metadata server for managing the appliance SAS environment
* Minimum Recommended hardware
Node 1
SAS High-Performance Analytics Environment
Root Node
SAS Hadoop/HDFS
Name Node2 x # Core x64 Processors## GB RAM * / Linux x64
SAS LASR Analytic Server
Distributed
Multiple Servers
Higher cost of entry
‘No’ limits to growth
‘Snap’ load and unload to HDFS from
memory
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS VERSION : SAS® VISUAL ANALYTICS VS. SAS® VAAR
VAAR
Limited functionality
No license cost
Usage tied to solution
VA
Full featured Visual Analytics
Standard per core pricing
Enterprise Usage
Or : Use BOTH!
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS METADATA : SHARED VS. SEPARATE
Shared Separate
What are the solutions being used by the end users Same Different
What release of the SAS foundation platform do the
solutions to be implemented use
Same Different
What are the requirements for High Availability, Disaster
Recovery
Same Different
Is it a common community, specific to solution, grouped by
security or privacy requirements
Same Different
What is the release / upgrade cadence for the software Same Different
Are there dependencies on other software Yes No
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILSHADOOP : SAS PROVIDED OR COMMERCIAL
DISTRIBUTION
SAS Provided : SAS High-
Performance Deployment for Hadoop
Integrated in install (SDM)
Limited functionality
No license cost
Support through SAS Technical Support
SAS High-Performance Deployment for Hadoop support.sas.com : Commercial Hadoop Distribution Support Matrix
Commercial Distribution
Separate installation
Full featured Hadoop environment
No license cost
Support through commercial vendor
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
Hadoop Cluster
HEADS OR TAILS VA LOCATION : ASYNCHRONOUS VS. CO-LOCATED
Asynchronous
Math and Memory Nodes
Separate performance impact
Use Embedded Processes to parallelize
SAS® LASR ANALYTIC SERVERS
Hadoop Cluster
Co-Located with Hadoop
Hadoop nodes
Cross performance impact
Snap-in, Snap-out HDFS persisted data
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS LASR SERVERS : DEFAULT (2) OR MORE
Default
Default Installation
All groups share the memory
Can still be secured through metadata
Private
Additions to default
May help restrict usage of memory by
group
May require slightly more administration
Secured through metadata
SAS® LASR ANALYTIC SERVERS
Default Public
SAS® LASR ANALYTIC SERVERS
Default Public Private1 Private2 PrivateX
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS WORKING WITH DATA : DATA BUILDER OR EXTERNAL
SAS® Visual Data Builder
End user self-service
Simple tool with focused set of capabilities
Users need only limited set of permissions to
the data
Minimum contact with SQL
Data created for a specific goal and shared
with a small team
External (e.g. SAS DI Studio, SAS EG)
IT owned and managed
Complex tool with wide range of capabilities
Need access to a larger set of enterprise data
Programming and scripting skills are
important
Data created for multiple goals and
shared across multiple divisions
/* PROC FREQ example */
/* Create a Hive table */
data myhdp.myUserID_class;
set sashelp.class;
run;
/* Run PROC FREQ on the class table */
proc freq data=myhdp.myUserID_class;
tables sex * age;
where age > 9;
title 'Frequency';
run;
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS DATA FORMATS : ABT OR START SCHEMA
ABT
+ Best Performance
+ Easy to append, update and delete rows
- Large memory footprint
- Hierarchy data item replication across
analytical base tables
Star Schema
+ Smaller memory footprint
+ Reuse multiple dimension tables across schema
views
- Performance overhead due to execution of joins at
action time
- Schema must be reinitialized to reflect appended,
updated, and/or deleted rows
ANALYTICAL BASE TABLES
The structure of an analytical base table is as follows:
• Flat, fully materialized (that is, pre-joined)
• Atomic (that is, at the lowest level of granularity for reporting)
• Dimensions, facts, and measure variables all in one table
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS
IT’S REALLY UP TO YOU!
The Choice :
Copyr i g ht © 2016, SAS Ins t i tu t e Inc . A l l r ights reser ve d .
HEADS OR TAILS BIBLIOGRAPHY / LINKS
• SAS Technical Papers » SAS Visual Analytics
• Reeling Them Back In—Keeping SAS® Visual Analytics Users Happy
• See Section on “More than One LASR Server?”
• Self-Service Data Management: SAS® Visual Data Builder
• See Section on “Design Principles” for Visual Data BUilder vs. External DI Tools