Learning Objectives: - Identify best practices to migrate from on-premise deployments to Amazon EMR - Understand hot to move data from HDFS to Amazon S3 - Use Amazon EC2 Spot instances to save costs Businesses are migrating their analytics, data processing (ETL), and data science workloads running on Apache Hadoop, Spark, and data warehouse appliances from on-premise deployments to Amazon EMR in order to save costs, increase availability, and improve performance. Amazon EMR is a managed service that lets you process and analyze extremely large data sets using the latest versions of over 15 open-source frameworks in the Apache Hadoop and Spark ecosystems. This tech talk will focus on identifying the components and workflows in your current environment and providing the best practices to migrate these workloads to Amazon EMR. We will explain how to move from HDFS to Amazon S3 as a durable storage layer, and how to lower costs with Amazon EC2 Spot instances and Auto Scaling. Additionally, we will go over common security recommendations and tuning tips to accelerate the time to production.

1. Migrating Big Data Workloads to
Amazon EMR
Anthony Nguyen
Senior Big Data Consultant (aanwin@amazon.com)
June 1st, 2017

2. Agenda
• Deconstructing current big data environments
• Identifying challenges with on-premises or unmanaged architectures
• Migrating components to Amazon EMR and AWS analytics services
- Choosing the right engine for the job
- Building out an architecture
- Architecting for cost and scalability
- Security
• Customer migration stories
• Q+A

3. Deconstructing current
big data environments

4. On-premises Hadoop clusters
• A cluster of 1U machine
• Typically 12 Cores, 32/64 GB
RAM, and 6 - 8 TB of HDD ($3-4K)
• Networking switches and racks
• Open-source distribution of
Hadoop or a fixed licensing term
by commercial distributions
• Different node roles
• HDFS uses local disk and is sized
for 3x data replication
Server rack 1
(20 nodes)
Server rack 2
(20 nodes)
Server rack N
(20 nodes)
Core

5. Workload types running on the same cluster
• Large Scale ETL: Apache Spark, Apache Hive with Apache Tez or
Apache Hadoop MapReduce
• Interactive Queries: Apache Impala, Spark SQL, Presto, Apache
Phoenix
• Machine Learning and Data Science: Spark ML, Apache Mahout
• NoSQL: Apache HBase
• Stream Processing: Apache Kafka, Spark Streaming, Apache Flink,
Apache NiFi, Apache Storm
• Search: Elasticsearch, Apache Solr
• Job Submission: Client Edge Node, Apache Oozie
• Data warehouses like Pivotal Greenplum or Teradata

6. Security
• Authentication: Kerberos with local KDC or
Active Directory, LDAP integration, local user
management, Apache Knox
• Authorization: Open-source native authZ (i.e.,
HiveServer2 authZ or HDFS ACLs), Apache
Ranger, Apache Sentry
• Encryption: local disk encryption with LUKS,
HDFS transparent-data encryption, in-flight
encryption for each framework (i.e., Hadoop
MapReduce encrypted shuffle)
• Configuration: different tools for management
based on vendor

7. Swim lane of jobs
Over utilized Under utilized

8. Role of a Hadoop administrator
• Management of the cluster (failures,
hardware replacement, restarting
services, expanding cluster)
• Configuration management
• Tuning of specific jobs or hardware
• Managing development and test
environments
• Backing up data and disaster recovery

9. Identifying challenges

10. Over utilization and idle capacity
• Tightly coupled compute and storage requires buying
excess capacity
• Can be over-utilized during peak hours and underutilized
at other times
• Results in high costs and low efficiency

11. Management difficulties
• Managing distributed applications and availability
• Durable storage and disaster recovery
• Adding new frameworks and doing upgrades
• Multiple environments
• Need team to manage cluster and procure hardware

12. Migrating workloads to
Amazon EMR

13. Why Amazon EMR?
Low Cost
Pay an hourly rate
Open-Source Variety
Latest versions of software
Managed
Spend less time monitoring
Secure
Easy to manage options
Flexible
Customize the cluster
Easy to Use
Launch a cluster in minutes

14. Key migration and TCO considerations
• DO NOT LIFT AND SHIFT
• Decouple storage and compute with S3
• Deconstruct workloads and map to open-source tools
• Transient clusters and auto scaling
• Choosing instance types and EC2 Spot Instances

15. Translate use cases to the right tools
Storage
S3 (EMRFS), HDFS
YARN
Cluster Resource Management
Batch
MapReduce
Interactive
Tez
In Memory
Spark
Applications
Hive, Pig, Spark SQL/Streaming/ML, Flink, Mahout, Sqoop
HBase/Phoenix
Presto
Athena
Streaming
Flink
- Low-latency SQL -> Athena or Presto or Amazon Redshift
- Data Warehouse / Reporting -> Spark or Hive or Glue or Amazon Redshift
- Management and monitoring -> EMR console or Ganglia metrics
- HDFS -> S3
- Notebooks -> Zeppelin Notebook or Jupyter (via bootstrap action)
- Query console -> Athena or Hue
- Security -> Ranger (CF template) or HiveServer2 or IAM roles
Glue
Amazon Redshift

16. Many storage layers to choose from
Amazon DynamoDB
Amazon RDS Amazon Kinesis
Amazon Redshift
Amazon S3
Amazon EMR

17. Decouple compute and storage by using S3
as your data layer
HDFS
S3 is designed for 11
9’s of durability and is
massively scalable
EC2 Instance
Memory
Amazon S3
Amazon EMR
Amazon EMR
Intermediates
stored on local
disk or HDFS
Local

18. HBase on S3 for scalable NoSQL

19. S3 tips: Partitions, compression, and file formats
• Avoid key names in lexicographical order
• Improve throughput and S3 list performance
• Use hashing/random prefixes or reverse the date-time
• Compress data set to minimize bandwidth from S3 to
EC2
• Make sure you use splittable compression or have each file
be the optimal size for parallelization on your cluster
• Columnar file formats like Parquet can give increased
performance on reads

20. TCO – Transient or long running clusters

21. Options to submit jobs
Amazon EMR
Step API
Submit a Spark
application
Amazon EMR
AWS Data Pipeline
Airflow, Luigi, or other
schedulers on EC2
Create a pipeline
to schedule job
submission or create
complex workflows
AWS Lambda
Use AWS Lambda to
submit applications to
EMR Step API or directly
to Spark on your cluster
Use Oozie on your
cluster to build
DAGs of jobs

22. Performance and hardware
• Transient or long running
• Instance types
• Cluster size
• Application settings
• File formats and S3 tuning
Master Node
r3.2xlarge
Slave Group - Core
c4.2xlarge
Slave Group – Task
m4.2xlarge (EC2 Spot)
Considerations

23. On-cluster UIs to quickly tune workloads
Manage applications
SQL editor, Workflow designer,
Metastore browser
Notebooks
Design and execute
queries and workloads

24. Spot for
task nodes
Up to 80%
off EC2
On-Demand
pricing
On-demand for
core nodes
Standard
Amazon EC2
pricing for
on-demand
capacity
Use Spot and Reserved Instances to lower costs
Meet SLA at predictable cost Exceed SLA at lower cost

25. Instance fleets for advanced Spot provisioning
Master Node Core Instance Fleet Task Instance Fleet
• Provision from a list of instance types with Spot and On-Demand
• Launch in the most optimal Availability Zone based on capacity/price
• Spot Block support

26. Lower costs with Auto Scaling

27. Security - Encryption

28. Security – Authentication and Authorization
Tag: user = MyUserIAM user: MyUser
EMR role
EC2 role
SSH key

29. Security - Authentication and Authorization
• LDAP for HiveServer2, Hue, Presto,
Zeppelin
• Kerberos for Spark, HBase, YARN,
Hive, and authenticated UIs
• EMRFS storage-based permissions
• SQL standards-based and storage-
based authorization
AWS Directory Service
Self-managed Directory

30. Security - Authentication and Authorization
• Plug-ins for Hive, HBase,
YARN, and HDFS
• Row-level authorization for Hive
(with data-masking)
• Full auditing capabilities with
embedded search
• Run Ranger on an edge node –
visit the AWS Big Data Blog
Apache Ranger

31. Security – Governance and Auditing
• AWS CloudTrail for EMR APIs
• S3 access logs for cluster S3 access
• YARN and application logs
• Ranger for UI for application level auditing

32. Customer Examples

33. Petabytes of data generated
on-premises, brought to AWS,
and stored in S3
Thousands of analytical
queries performed on EMR
and Amazon Redshift.
Stringent security requirements
met by leveraging VPC, VPN,
encryption at-rest and in-
transit, CloudTrail, and
database auditing
Flexible
Interactive
Queries
Predefined
Queries
Surveillance
Analytics
Data Management
Data Movement
Data Registration
Version Management
Amazon S3
Web Applications
Analysts; Regulators
FINRA: Migrating from on-prem to AWS

34. Lower Cost and Higher Scale than On-Premises

35. Learn
Models
ModelsImpressions
Clicks
Activities
Calibrate
Evaluate
Real
Time
Bidding
S3
ETL Attribution
Machine
Learning
S3Amazon
Kinesis
• 2 Petabytes Processed Daily
• 2 Million Bid Decisions Per Second
• Runs 24 X 7 on 5 Continents
• Thousands of ML Models
Trained per Day

36. FINRA saved 60% by moving to HBase on EMR

37. Anthony Nguyen
aanwin@amazon.com
aws.amazon.com/emr
blogs.aws.amazon.com/bigdata
Q+A Thank
you!