Part 2: Cloudera’s Operational Database: Unlocking New Benefits in the Cloud

1© Cloudera, Inc. All rights reserved.
Operational Database
in the Public Cloud
Ryan Lippert | Cloudera Product Marketing

What’s Driving Operations to the Cloud?
● Increased Agility: End-user self-service
● Elasticity: Optimize infrastructure usage
● Lower Overall TCO
● Executive Mandate: Minimize on-
premises datacenter footprint
Big data deployments in cloud are accelerating; why?

Overview
Cloudera’s Operational Database

Build Data-Driven Applications to Deliver Real-Time Insights
Operational Database
Attributes
Fast
• Real time model serving w/ <15ms latency
• Limitless concurrency (>100M updates/sec)
Secure
• Native encryption
• Audit
Easy
• Stream ingest, processing, NoSQL, and real-time
analytics together
• Best-in-class management and cloud automation

Visualization
Processing/
Exploration
Storage
Unique Components
Fast/random reads and writes
via a high-performance,
distributed NoSQL data store
HBase
Fast analytics on fast data
with a relational structure
Kudu
Integration with the leading BI
tools
BI Partners
Faceted, text-based search for
data exploration and
democratization
Cloudera
Search
Powerful and flexible
processing, streaming, and
SQL
Spark
Multi-Storage
Multi-Environment
Encryption, Key Trustee
Navigator
Storage & Governance

Operational
Database
Durable, low latency storage for web
applications, message stores, and
mission critical operational activities.
Web-Scale Data Depot
Identifying meaningful events
based on multiple data streams
and taking action.
Complex Event Processing
Use data and current/past
events to score and serve the
likelihood of subsequent
events.
Model Scoring/Serving

Web Scale Data Depot
Key Applications for Operational Database
Real-Time Data Access
Low-latency and high concurrency enable broad-based
access to real-time information, yielding informed
decisions
Enterprise Data Apps
Build company-wide, easy-to-use apps to enable
employees or customers to interact with pertinent data
IoT Data Ingestion and Collection
Ingest, process, and serve IoT data in real time to take
advantage of instrumentation investments
Web-Scale Data Management
Store information from broad sets of customer
interaction occurring online, in-app, or in-store
Ingests over 4 million
homes worth of
energy data, and
provides reports that
help customers save
millions
Serves real-time
market data for over
40M instruments;
ingests 2.5M
transactions/second,
serves 3.5M
messages/second

Complex Event Processing
Cybersecurity and
Advanced Persistent Threats (APT)
Protect data with data; by keeping full-fidelity records
of network activity, anomalies can be surfaced and
thwarted
Network Health Monitoring
Maintain performance within an enterprise network by
identifying and remedying problems in real-time
IoT Predictive Maintenance
Use IoT sensor data from an unlimited number of
sources to proactively predict and fix problems with
physical equipment
Reduced detection of
APT from hundreds of
days to minutes; now
scales to thousands of
endpoints vs.
hundreds
Remote diagnostics
IoT platform reduces
fleet maintenance on
180,000+ vehicles by
30-40%

Model Scoring & Serving
Cross-Sell/Up-Sell & Personalization
Leverage a long history of purchases among a broad
population to create personalized offers, in real-time,
that are likely to be actioned by shoppers
Fraud Prevention
Compare recent financial transactions/claims with a
company-wide history of nefarious transactions to
identify and prevent fraud in real-time
Customer Profitability
Quickly identify high-value customers via individual
characteristics that correlate with profitability; focus
acquisition and retention on these segments
Lower cart
abandonment; 3x
higher open email
rate; decreased
bounce rate by 20%;
time to update
indexes from a day to
15 min
Can now provide
customers with 300-
400% higher CTR, 10x
more return visits, and
longer sessions

Benefits of the Public Cloud

What’s Driving Operations to the Cloud?
● Increased Agility: End-user self-service
● Elasticity: Optimize infrastructure usage
● Lower overall TCO
● Executive Mandate: Minimize on-
premises datacenter footprint
Big data deployments in cloud are accelerating; why?

Advantages of Our Approach
Cloud-Native & On-Premises
Go Beyond SQL
• Open Architecture: Open
formats and open storage
• Shared data across SQL and
non-SQL workloads
Data Flexibility
• Faster, more agile data
acquisition
• Data portability: Open
formats and open storage
Cost-Effective Scalability
• Elastic scale on-prem or in
the cloud
• Cloud-native pay-per-use
and transience
• Proven at big data scale
Hybrid
• Runs across multi-cloud &
on-prem
• Multi-storage over S3, HDFS,
Kudu, Isilon, DSSD, etcShared Data

Operational Database in the Cloud
Public Cloud Benefits
Cost
Considerations
• Low-cost backup and
disaster recovery
• Development and
testing environments
easy to deploy and
decommission
Convenience
Considerations
• Elastic growth for
tightly provisioned
workloads makes
expansion easy, and
enables a lower-cost
steady state
• Fast and easy
provisioning of
additional clusters
helps projects move
quickly

Operational Database Cloud Architecture
Applications
Long lived
Prod Cluster
Operational
DB
Director
Provision
CM
Manage/Provision
Operational
DB
Temporary
Dev/Test Cluster
Burst Batch
Processing
Data copy for burst processing
or read/write temp clusters
1
Easy
Provisioning
2
Dev/Test
Provisioning
3
Burst processing of
large amounts of data
4
Low cost
backups
Data
Sources Spark
Streaming
S31
EBS2
AWS
Infrastructure
Benefits
1. S3, Azure Blob Storage, etc.
2. EBS, Azure Premium Storage, etc.

Easy Provisioning of New Clusters
Operational Database in the Public Cloud
Easy
Provisioning
Business Challenge
• On-premises installations can be slow to roll out, particularly for PoC
engagements with long procurement cycles
• Some organizations can take 3-6 months to execute this process,
slowing developers and creating anchors to legacy technology
Cloud-Enabled Solution
• Cloudera enables customers to go to the public cloud with industry-
leading software
• Workloads can be moved across clouds or to on premise clusters,
preventing cloud lock-in
Details
• Cloudera provides the ability to quickly provision a new cluster for
operational use cases in the public cloud
• Rapid provisioning without the permanent cost of internal
infrastructure helps make Cloudera a fast and easy choice for PoC’s
• Companies with Cloudera-trained employees have the ability to test
and prove/disprove new use cases quickly, delivering more value to
business

Easy Provisioning of New Clusters
Application
Operational
DB
Cloud Instances
Director
Provision
Cloud Storage
Application
Cloud Instances
Direct Attached Storage
Fast Cloud
Storage1
Operational
DB
Instance
Storage
Director
Provision

Spark Streaming and Operational DB in the cloud
For real-time processing and serving architectures
Availability Zone Applications
Ingest
Streaming
Data
Spark Streaming
running on a
dedicated
permanent cluster
Spark
Streaming
Operational
DB
Operational DB on a
dedicated permanent
cluster
Both clusters in the
same availability
zone
Deliver/
Serve
Data

Creating Dev and Test Environments
Development
and Testing
Environments
Business Challenge
• Development and testing environments are expensive to
maintain/configure, difficult to secure with real data, and have
different projects competing for a finite pool of resources
• Cloudera in the public cloud enables development and testing
environments to be provisioned quickly, securely, and for the
required period of time
Details
• Public cloud offerings with Cloudera enable the ability to easily and
quickly replicate a production instance of data to a testing
environment
• Test environments can be configured with all the security of the
production, without the risk of overloading critical infrastructure
• Temporary instances mean environments are purpose-created,
time-bound, and less competition for test/dev resources

Creating Development and Testing Environments
Fast, Easy, and Secure Development & Testing in the Cloud
Production
Ready
Data
Application
Production
Instance Delivers
Data to Users
and Applications
Cloud
Object
Storage1
Fast Cloud
Storage2
Develop-
ment and
Testing
Production Environment Dev/Test Environment
Secure Dev/Test
Environment

Burst Provisioning for ETL
Leverage Cloud
for ETL
Business Challenge
• ETL is a difficult process that consumes a large amount of resources
and can create bottlenecks depending on the nature of batch
processes or data spikes
• Cloudera can leverage the elasticity of resources in the public cloud
to help businesses handle large ETL jobs, regardless of whether they
are anticipated or not
Details
• Unexpected surges in traffic, regular batch jobs that are growing in
size, and other high-volume data ingestion issues can create
bottlenecks that slow insight into the business; standard on-
premises ETL may have a difficult time recovering from the surge,
resulting in lost data
• Public cloud instances of Cloudera enable additional ETL resources
to be added temporarily, overcoming the deluge

Burst Provisioning for ETL
Keeping your Operational Database Real-Time in the Cloud
Operational
DBData
Surge
1
Data
Sources
Cloud
StorageData Pushed
to Cloud
Storage
2
Data to
Cloud
Instances
for Batch
Processing
3
AWS EBS
Instances
4
Transformed
Data Returned
to Cloud
Storage
5 Transformed Data
Sent to HBase
Application
6 Data Served to Application

Backup and Disaster Recovery
Backup and
Disaster Recovery
Business Challenge
• Businesses struggle to create backup of their critical data, including
challenges with geographical dispersion, maintenance costs,
frequency of backup, etc.
• Public clouds offer the ability to take frequent snapshots of the data
within your CDH cluster
Details
• By snapshotting to a remote public cloud datacenter, companies can
take advantage of geographically dispersed data copies
• Cheap storage enables more frequent backups, enabling a more
recent copy of data to be recovered in case of problems
• Navigate issues associated with data sovereignty

Backup and Disaster Recovery
Reduce Risk by Backing-Up Operational Database Data in the Public Cloud
On-Premises
Instance
Cloud Object
Storage1
Cloud
Instance
Snapshot
of Data
1
Cloud Object
Storage1
Restore
2
Snapshot
of Data
1
Restore
2

CDH HBase on EBS vs. EMR HBase on S3
EMR and S3 options for HBase
- Amazon can run HBase on S3 and get consistency via
a proprietary EMR-FS connector
- Cloudera leverages EBS for HBase cloud deployments
Customer aim: performance and price
- S3 saves on storage costs relative to EBS, but
increases compute costs as you need more EC2
instances to get to the same metrics
- So, for S3 storage costs go down, but compute costs
go up; few use cases can combine low cost on both
axises
Qualitative customer concerns
- From Cloudera: High availability, automated
configuration, manageability (monitoring/alerts),
support (from Cloudera’s HBase developers)

Instance Recommendations
Easy Provisioning of always on clusters:
Model vCPU Mem(GiB) Storage(GB)
d2.xlarge 4 30.5 3 x 2000 HDD
d2.2xlarge 8 61 6 x 2000 HDD
d2.4xlarge 16 122 12 x 2000 HDD
d2.8xlarge 36 244 24 x 2000 HDD
Data Nodes:
Master Nodes:
Model vCPU Mem(GiB) Storage(GB)
c3.8xlarge 32 60 2 x 320 SSD
Snapshot Backups: S3
Smaller versions are
recommended to stagger the
impact of full block reports and
garbage collection.
The master node memory
should be sized inline with the
cluster size, c3.xlarge supports
very large cluster sizes but
smaller master nodes are
possible.

Transient clusters with permanent storage using Director:
Model vCPU Mem(GiB) Storage
C4.large 4 30.5 EBS (4000 Mbps dedicated)
Data Nodes:
Master Nodes:
Model vCPU Mem(GiB) Storage (GB)
c3.8xlarge 32 60 2 x 320 SSD
Storage, throughput workloads (ETL, etc.):
Volume Type Volume Size IOPS Throughput
st1 500 GiB – 16 TiB 500 800 MiB/s
Volume Type Volume Size IOPS Throughput
io1 4 GiB – 16 TiB 20,000 800 MiB/s
Storage, real time workloads (HBase, etc.):
These are default
recommendations; they will
vary based on the specifics of
each use case.
We recommend deploying
more than the lowest tier, SC1,
as the throttling limits are
reached quickly, which brings
down the database.

Always on Spark Streaming cluster
• Spark Clusters have homogenous nodes i.e. no special “master” node
m4.2xlarge 8 32 EBS (1000 Mbps dedicated)
Default:
m3.2xlarge 8 61 160GB SSD
Very Memory Intensive Workloads:
Best balance of memory and
compute
Examples are workloads that
cache RDDs/Dataframes or
maintain in-memory state via the
updateStateByKey(…) function.
c4.2xlarge 8 15 EBS (1000 Mbps dedicated)
Very Compute Intensive Workloads:
Examples are workloads that may
perform compute intensive
machine learning operations to
score incoming events.

• Get Started with Cloudera in the Cloud:
• www.cloudera.com/downloads
• Learn more about Cloudera’s Operational DB:
• https://www.cloudera.com/solutions/operational-database.html
• Learn about Data Engineering Workloads in the Cloud:
• www.cloudera.com/about-cloudera/events/webinars/cloud-webinar-series.html
Next Steps

Thank You

Part 2: Cloudera’s Operational Database: Unlocking New Benefits in the Cloud

Recommended

More Related Content

What's hot (20)

Viewers also liked (12)

Similar to Part 2: Cloudera’s Operational Database: Unlocking New Benefits in the Cloud (20)

More from Cloudera, Inc. (20)

Recently uploaded (20)

Part 2: Cloudera’s Operational Database: Unlocking New Benefits in the Cloud