Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
13433 - When Decisions Can’t Wait: From Analysis to Action – in Real Time
Dan Soceanu, Senior Solutions ArchitectSAS
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
SESSION AGENDA
4Use Cases
1
2
3
The Need for Streaming Analytics
The Streaming Analytics Lifecycle
SAS® Event Stream Processing
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
BIG DATA NEW LANDSCAPE – NEW NEEDS
Volume
Velocity
Variety
Immediate low latency answers
Reduced time to decision action
Continuously evaluate opportunities and risks
More agile, more responsive
Better equipped to address big data
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
STREAMING ANALYTICS DEFINED
… it is about applying analytics while the data is in motion, before
it is stored – and keeping what it is relevant
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
Event = a message indicating that something has
happened
Event Stream = an ordered sequence of events, of the same type
ESPEvent Streaming Processing = matching and transforming source events into result events;Data-in-Motion analyses data before storage
DEFINITIONS WHAT IS AN EVENT, EVENT STREAM AND ESP?
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
Batch-Loaded Data
WarehouseMicro-Batch
Data Warehouse
Trickle-feed DW with CDC
Complex Event Processing
Event Stream Processing
Days Hours Minutes Seconds Milliseconds Microseconds
NEW ERA OF INFORMATION PROCESSING
THE EVOLUTION OF OPERATIONALIZING ANALYTICS
Move analysis to
event source,
Analyze before
data is stored,
Keep what is relevant.
High-speed querying of
data in streams, and
applying algorithms to
the event data.
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
NEW ERA OF INFORMATION PROCESSING
USERS NEED IMMEDIATE DECISIONS
Processing streaming data is about getting immediate answers to reduce time to decision
Time to decisionMicro-seconds Days
Batch
Streaming
Streaming
Analytics
Micro-Batch
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
STREAMING ANALYTICS
CROSS-INDUSTRY APPLICABILITY AND VALUE
Cyber Security
IT Operations
Real Time Marketing
Supply Chain
Fraud Detection
Capital Markets
Manufacturing
Industry, Energy
Enterprise Decisions
Telecommunications
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
WHEN TO USE SAS ESP?
Digital Security
Connected Devices (IOT) & Sensor Networks
On-line Behavior
When milliseconds matter Where latency devalues events (e.g., operational data) When volumes (i.e. throughput rate) overwhelm existing
analytics When velocity leads to unacceptable latency
SAS EVENT STREAM PROCESSING
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
BIG DATA TRADITIONAL ANALYTICS LIFECYCLE
DeployETL
Data Data Storage
f
Access – Store - Analyze
Alerts / Reports
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
SENSE – UNDERSTAND - ACT STREAMING ANALYTICS LIFECYCLE
DeployETL
Data Data Storage
Alerts / Reports/ Decisioning
De
plo
y
f
Streaming Data Intelligent Filter / Transform
Streaming Model Execution
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
DISTRIBUTED ANALYTICS STREAMING ANALYTICS VALUE STREAM
DeployETL
Data Data Storage
Alerts / Reports/ Decisioning
De
plo
y
f
IoT Data Intelligent Filter / Transform
Streaming Model
Execution
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
STREAMING ANALYTICS CONTINUUM
Cloud Streaming Edge
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
SAS®
Event Stream ProcessingOverview
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
PURPOSE-BUILT PROVIDES REAL-TIME ANSWERS
SAS Event Stream Processing
Processing high throughput, low latency streaming events requires moving from
(reactive) real-time to (proactive) real-time
Reactive real-time Proactive real-time
High
throughput
Low latency
Medium
throughput
and latency
Continuously analyze to define relevant
action
Real-time action occurs as the result of pattern
detection
SAS Real-Time Solutions
Listen and react to incoming requests
Real-time action occurs as the result of a triggering event
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
Known Patterns
•Rule: Set up rules for filter fraudulent transactions
•Examples: two transactions in different time zones within short period of time
Unknown Patterns
•Anomaly Detection: Detect individual and aggregate abnormal patterns
•Examples: Mean , standard deviation, percentiles, univariate and multivariate regression, clustering, sequence analysis, peer group analysis
Complex Patterns
•Advanced Analytics: Perform knowledge discovery, data mining, predictive assessment
•Examples: Neural networks, decision trees, generalized linear models, econometric models, gradient boosting
Associative Link Patterns
•Social Network Analysis: Perform knowledge discovery through associative linkage analysis
•Examples: Social network + linkage analysis + community detection + advanced analytics
PATTERN DETECTIONPURPOSE-BUILT
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
PURPOSE-BUILT IDENTIFIES PATTERNS OF INTEREST
One or more:
• Pre-built data quality routines
• Business rules and policy definitions
• Advanced analytics:
• Scoring events (models
developed on data at rest)
• Machine Learning clusters
(models defined in-stream)
• Extract entities, classify and
identify sentiment (NLP methods)
• Filter, aggregate and correlate events
• Reference historic data (store in-
memory)
• Continuous queries or periodic queries
• Pattern detection at event stream source
• Offline, data at rest identifies emerging trends
• Feed new insights back into event streams
• Dynamically update queries into live stream
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
PURPOSE-BUILT FAST AND ADAPTIVE ACTION
SAS-generated Insights
Event Actions
SAS In-Memory
SAS®
Event Stream Processing Model
Continuous Query
Pu
bli
sh
Su
bs
cri
be
Streaming Events
Enrichment Data
Analytic Models
Business Rules
Pattern detection at event stream source
Offline, data at rest identifies emerging trends
Feed new insights back into event streams
Dynamically update queries into live stream
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
PURPOSE-BUILT A SINGLE LEARNING ENVIRONMENT
Dataflow centric modeling
Drag & drop visual
modeler
Visual, XML or C modeling
Dynamic model update
Publish & Subscribe API
(Java, C, Python)
Model definition and maintenance, simplified with visual
modeling interfaceo Create and maintain streaming models easily for fast and flexible
adaptive actiono Full set of components to build any type of processo Incremental model testing
Easy deployment of streaming analytic models
o Deployment of existing analytic models using embedded SAS® DS2, SAS® Datastep or Python code
o Deploy ESP models as XML fileso Dynamic model updates
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
PURPOSE-BUILT ENGINEERED FOR SPEED
Throughput - how many events per second can be ingested
Latency - the time it takes for an event to be processed through the defined workflow
• Millions of events per second throughput
• Millisecond-microsecond response latency
• On standard commodity hardware
Event Streams are high throughput, low latency data flows
SAS Event Stream Processing provides:
Continuous in-memory
processing
OS native application
Threaded pool - clustering
Linear scalability
Fastest ESP in the market
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
PURPOSE-BUILT ENGINEERED FOR AGILITY
Lightweight embedding
technology
Cloud ready
OS native application
Clustering
Dynamic model update
Low footprint OS native application
From lightweight embedded technology to cloud
distributed architecture
Fulfill new architecture needs
Edge Small Large Cluster Cloud
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
STREAMING ANALYTICS COMPLEMENTS SAS ANALYTICS
Real Time Alerts
VISUALIZATION, ALERTS
Continuous,Incremental Analytics
ADVANCED ANALYTICS
HIGH PERFORMANCE
ANALYTIC SOLUTIONS
Risk Analytics
Fraud Analytics
Asset Performance Analytics
Customer Intelligence
Decision Management
Visual Analytics
Etc …
STREAMING DATA ANALYSIS
EVENT STREAM
PROCESSING
ENGINE
Continuous processing of
events, in high-volume streams,
to detect actionable information
Rules
Correlation
BIG DATA STREAMS
BATCH DATA ANALYSIS
ACCESS
ENGINES
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
IN-STREAM ANALYTICS PROCEDURAL WINDOW
Build analytical models using SAS® DS2
• Decision Tree
• Neural Network
• Regression
• Rule Induction
• Scoring
• And more
Or SAS DATAStep, C/C++, Python
• Coming soon: R
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
ECOSYSTEM INTEGRATION 300+ ENDPOINTS
OPEN SOURCE
SYSTEMS & APPLICATIONS
PUBLISH & SUBSCRIBE API
CONNECT TO ANY SYSTEM WITH JAVA, C, PYTHON
FULLY DOCUMENTED AND EASY TO USE
RendezVous
STANDARDS
FILE/SOCKET
XML / JSON
ODBC
JMS
MQTT
SYSLOG
DB LOG SNIFFERS
HTTP RESTFUL
SMTP
NETWORK SNIFFERS
WEB SERVICES
* *
*
*
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
SAS®
Event Stream Processing Use Cases
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
REAL-TIME CONTEXTUAL DATA MINING
BUSINESS ISSUE
• Needed a solution for real-time campaign management
(anytime, anywhere market challenges)
• Current process was extremely manual and resulted in high-
latency reporting
• Lacked advanced analytical capabilities and was limited in
terms of real-time capabilities
RESULTS
• Contextual real-time analysis of the streaming call data
records (CDRs) - 20,000+ requests per second
• The right offer at the right time with the right channel
• Enhanced the accuracy of predictions and decisions
COMMUNICATIONS
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
PATIENT MANAGEMENT
BUSINESS ISSUE
• Detect relevant patterns in patient real-time data to alert critical
care teams
• Address Alert Fatigue٭
• Patient vital statistics from various sensors across different
equipment
• Incoming lab results joined with real time sensor data
RESULTS
• Monitor data to trigger actions based upon detected patterns
• Send messages across email and SMS
• Alert immediately appropriate critical care teams
• Send immediate recommendation to remote patient
HEALTH CARE
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
ONLINE FRAUD DETECTION BANKING
BUSINESS CHALLENGE
• Under attack from fraudsters
• Bank wanted to drastically quicken its reporting capabilities and be
able to move from an overnight to intraday reporting speed
• Lack of key elements in detection logic (e.g., beneficiary profiles,
PC session logs)
RESULTS
As a result of this delivery, the bank's market risk managers are now
able to:
• Access 10 times the volume of data on a given day
• Receive risk data four hours earlier than before
• Analyze large data sets of up to seven terabytes of granular risk
data on demand
• Access data through a central feature-rich, web-based portal
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
SAS EVENT STREAM PROCESSING
KEY TAKEAWAYS
Detect and monitor
events continuously, taking real-time relevant
action for greatest impact
REAL-TIME
RELEVANT
ACTION
Retain only what’s appropriate, and filter
and cleanse before big data is stored
FOCUS ON
RELEVANT
DATA
One managed, easy to use
environment to examine, assess, action and improve
streaming analytics
GOVERNED
MULTI-PHASE
ANALYTICS
#analyticsx
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
Questions?Thank you for your time today!
C o p y r ig ht © 201 6, SAS In st i tute In c. A l l r ig hts r ese rve d.
#AnalyticsX