Hadoop Summit - 2014
Cost of Ownership for Hadoop
Implementation
Santosh Jha,
Steve Ackley
Part 1 – Estimating TCO
Iceberg
Estimating TCO is hard.
Like an iceberg, many
costs are hidden.
Example :
integration of Big Data
within the existing
ecosystem.
Hadoop Implementations
Hadoop deployment methods
Sample Vendors
Hortonworks IBM, EMC AWS EMR
Cloudera Oracle, Teradata Rackspace Altiscale
MAPR VMware Gogrid Quoble
On Premise
Hadoop
Appliance
Hadoop
Hosting
Hadoop as a
service
Bare Metal Cloud
On-Premise Cost Categories
Cost Group Item
Hardware/Infrastructure Costs Servers , Peripherals, Network
Storage
Communication Costs Local Area Network , Wide Area Network
Remote Access
Software Costs License/Subscription Fees
Implementation Costs Development/customization/integration
Training , Consulting , Non Functional
Testing(Performance, Capacity, Security etc.)
Management Costs Hardware & software upgrades , Hardware &
software administration, Legal Cost
Support Costs Support staff, Staff training, Travel, Support
contracts, Overhead labor, High Availability Cost
Disaster Recovery Cost, Ticketing & Trouble
Shooting Cost, Monitoring Cost, Internal Audit Cost
Managing Risk
Cost Group Item
Vendor Vendor Viability
Control on Technical Architecture
Data Protection
Loss of Intellectual Property
Loss of Privacy
Internal IT Vendor Viability
Control on Technical Architecture
Data Protection
Loss of Intellectual Property
Loss of Privacy
Sample calculation
Inputs
Average Monthly HDFS (TB) 1500
Peak HDFS over Monthly (TB) 100
Monthly HDFS Growth (TB) 20
Average Monthly Compute ('000 SH) 20
Peak Compute (SH) 1400
Planning Cycle (Months) 36
Purchased Distribution No
Hadoop Admin Costs Included
Data from S3 Yes
Results without considering risk
0
1,000,000
2,000,000
3,000,000
4,000,000
5,000,000
6,000,000
7,000,000
8,000,000
Hadoop as a
service
On Premise Amazon EMR Hadoop
Distribution
on EC2
Cost over 36 Months
Cost over 36 Months
Managing Risk (Vendor) – Sample data
Managing Risk Risk Factor Weight(%) Calculated Risk
Vendor Viability 2 40 0.8
Control on Technical
Architecture 1 20 0.2
Data Protection 2 15 0.3
Loss of Intellectual
Property 1 10 0.1
Loss of Privacy 2 15 0.3
Total 1.7
Vendor Viability 1 - No Risk, 5 - Very High Risk with vendor viability
Control on Technical Architecture 1 - No Need to Control, 5 - Compelling Need to control technical architecture.
Data Protection 1 - High data protection provided by architecture and process, 5 - No data protection
Loss of Intellectual Property 1 - No IP, 5 - High business impact with the loss of IP
Loss of Privacy 1 - No privacy issue for the solution, 5 - High business impact with loss of Data
Managing Risk (Internal IT – Sample data)
Managing Risk Risk Factor Weight(%) Calculated Risk
Vendor Viability 1 40 0.4
Control on Technical
Architecture 1 20 0.2
Data Protection 2 15 0.3
Loss of Intellectual
Property 1 10 0.1
Loss of Privacy 2 15 0.3
Total 1.3
Vendor Viability 1 - No Risk, 5 - Very High Risk with vendor viability
Control on Technical Architecture 1 - No Need to Control, 5 - Compelling Need to control technical architecture.
Data Protection 1 - High data protection provided by architecture and process, 5 - No data protection
Loss of Intellectual Property 1 - No IP, 5 - High business impact with the loss of IP
Loss of Privacy 1 - No privacy issue for the solution, 5 - High business impact with loss of Data
Results after considering risk
0
2000000
4000000
6000000
8000000
10000000
12000000
14000000
Hadoop as a
service
On Premise Amazon EMR Hadoop
Distribution
on EC2
Cost over 36 Months
Cost over 36 Months
Part 2 - Deployment
Considerations
On-Premise Implementation – When?
• Well-defined use cases with a demonstrated ROI
• Developed and tuned Hadoop applications
• IT team with experience and bandwidth to
manage/maintain Hadoop and integrated
hardware/software stack - as well as troubleshoot job
problems
• Sufficient # of Nodes to Support:
o Growth in Data Sets
o “Bursty” Nature of Jobs
On-Premise Implementation – Company Profile
• Large enterprise with a strategic need for Big Data
Analytics
• Moved from an exploratory stage to enterprise
adoption
• Committed IT resources to support Hadoop
hardware/software stack
Hadoop as a Service – The Continuum
• Vendors manage the hardware
• Vendors install hadoop
• Vendors manage hadoop
Vendors Manage The Hardware
For Organizations that:
• Want to create a small cluster for a relatively
short period of time, for training and software
development purposes.
• Have a short-term processing need and no
internal capacity to support it.
• Do not have an IT organization that can install,
manage, maintain and operate the Hadoop
hardware/software stack, and can fix “broken”
jobs.
Vendors Install Hadoop
For Organizations that:
• Have a short-term need or small-scale Hadoop
requirement.
• Have Hadoop applications that are “bursty.”
• Have an IT organization that can operate the
Hadoop hardware/software stack, can manage
scaling the cluster, and can fix “broken” jobs.
• Do not need to tailor the hardware to their
specific requirements.
Vendors Manage Hadoop
For Organizations that:
• Do not have the IT organization that can install,
manage, maintain and operate the Hadoop
hardware/software stack, and fix “broken” jobs.
• Do not have the IT hardware infrastructure that’s
required.
• May need an “always on” Hadoop environment.
• Need service providers that:
• Can handle all aspects of the IT support for Hadoop.
• Can provide comprehensive SLAs.
• May offer hardware optimized for Hadoop.
19
Thank You
Contact :
steve@altiscale.com
Santosh.jha@aziksa.com

Cost of Ownership for Hadoop Implementation - Hadoop Summit 2014

  • 1.
    Hadoop Summit -2014 Cost of Ownership for Hadoop Implementation Santosh Jha, Steve Ackley
  • 2.
    Part 1 –Estimating TCO
  • 3.
    Iceberg Estimating TCO ishard. Like an iceberg, many costs are hidden. Example : integration of Big Data within the existing ecosystem.
  • 4.
    Hadoop Implementations Hadoop deploymentmethods Sample Vendors Hortonworks IBM, EMC AWS EMR Cloudera Oracle, Teradata Rackspace Altiscale MAPR VMware Gogrid Quoble On Premise Hadoop Appliance Hadoop Hosting Hadoop as a service Bare Metal Cloud
  • 5.
    On-Premise Cost Categories CostGroup Item Hardware/Infrastructure Costs Servers , Peripherals, Network Storage Communication Costs Local Area Network , Wide Area Network Remote Access Software Costs License/Subscription Fees Implementation Costs Development/customization/integration Training , Consulting , Non Functional Testing(Performance, Capacity, Security etc.) Management Costs Hardware & software upgrades , Hardware & software administration, Legal Cost Support Costs Support staff, Staff training, Travel, Support contracts, Overhead labor, High Availability Cost Disaster Recovery Cost, Ticketing & Trouble Shooting Cost, Monitoring Cost, Internal Audit Cost
  • 6.
    Managing Risk Cost GroupItem Vendor Vendor Viability Control on Technical Architecture Data Protection Loss of Intellectual Property Loss of Privacy Internal IT Vendor Viability Control on Technical Architecture Data Protection Loss of Intellectual Property Loss of Privacy
  • 7.
    Sample calculation Inputs Average MonthlyHDFS (TB) 1500 Peak HDFS over Monthly (TB) 100 Monthly HDFS Growth (TB) 20 Average Monthly Compute ('000 SH) 20 Peak Compute (SH) 1400 Planning Cycle (Months) 36 Purchased Distribution No Hadoop Admin Costs Included Data from S3 Yes
  • 8.
    Results without consideringrisk 0 1,000,000 2,000,000 3,000,000 4,000,000 5,000,000 6,000,000 7,000,000 8,000,000 Hadoop as a service On Premise Amazon EMR Hadoop Distribution on EC2 Cost over 36 Months Cost over 36 Months
  • 9.
    Managing Risk (Vendor)– Sample data Managing Risk Risk Factor Weight(%) Calculated Risk Vendor Viability 2 40 0.8 Control on Technical Architecture 1 20 0.2 Data Protection 2 15 0.3 Loss of Intellectual Property 1 10 0.1 Loss of Privacy 2 15 0.3 Total 1.7 Vendor Viability 1 - No Risk, 5 - Very High Risk with vendor viability Control on Technical Architecture 1 - No Need to Control, 5 - Compelling Need to control technical architecture. Data Protection 1 - High data protection provided by architecture and process, 5 - No data protection Loss of Intellectual Property 1 - No IP, 5 - High business impact with the loss of IP Loss of Privacy 1 - No privacy issue for the solution, 5 - High business impact with loss of Data
  • 10.
    Managing Risk (InternalIT – Sample data) Managing Risk Risk Factor Weight(%) Calculated Risk Vendor Viability 1 40 0.4 Control on Technical Architecture 1 20 0.2 Data Protection 2 15 0.3 Loss of Intellectual Property 1 10 0.1 Loss of Privacy 2 15 0.3 Total 1.3 Vendor Viability 1 - No Risk, 5 - Very High Risk with vendor viability Control on Technical Architecture 1 - No Need to Control, 5 - Compelling Need to control technical architecture. Data Protection 1 - High data protection provided by architecture and process, 5 - No data protection Loss of Intellectual Property 1 - No IP, 5 - High business impact with the loss of IP Loss of Privacy 1 - No privacy issue for the solution, 5 - High business impact with loss of Data
  • 11.
    Results after consideringrisk 0 2000000 4000000 6000000 8000000 10000000 12000000 14000000 Hadoop as a service On Premise Amazon EMR Hadoop Distribution on EC2 Cost over 36 Months Cost over 36 Months
  • 12.
    Part 2 -Deployment Considerations
  • 13.
    On-Premise Implementation –When? • Well-defined use cases with a demonstrated ROI • Developed and tuned Hadoop applications • IT team with experience and bandwidth to manage/maintain Hadoop and integrated hardware/software stack - as well as troubleshoot job problems • Sufficient # of Nodes to Support: o Growth in Data Sets o “Bursty” Nature of Jobs
  • 14.
    On-Premise Implementation –Company Profile • Large enterprise with a strategic need for Big Data Analytics • Moved from an exploratory stage to enterprise adoption • Committed IT resources to support Hadoop hardware/software stack
  • 15.
    Hadoop as aService – The Continuum • Vendors manage the hardware • Vendors install hadoop • Vendors manage hadoop
  • 16.
    Vendors Manage TheHardware For Organizations that: • Want to create a small cluster for a relatively short period of time, for training and software development purposes. • Have a short-term processing need and no internal capacity to support it. • Do not have an IT organization that can install, manage, maintain and operate the Hadoop hardware/software stack, and can fix “broken” jobs.
  • 17.
    Vendors Install Hadoop ForOrganizations that: • Have a short-term need or small-scale Hadoop requirement. • Have Hadoop applications that are “bursty.” • Have an IT organization that can operate the Hadoop hardware/software stack, can manage scaling the cluster, and can fix “broken” jobs. • Do not need to tailor the hardware to their specific requirements.
  • 18.
    Vendors Manage Hadoop ForOrganizations that: • Do not have the IT organization that can install, manage, maintain and operate the Hadoop hardware/software stack, and fix “broken” jobs. • Do not have the IT hardware infrastructure that’s required. • May need an “always on” Hadoop environment. • Need service providers that: • Can handle all aspects of the IT support for Hadoop. • Can provide comprehensive SLAs. • May offer hardware optimized for Hadoop.
  • 19.

Editor's Notes

  • #2 Welcome to Hadoop Summit 2014.
  • #3 Welcome to Hadoop Summit 2014.
  • #4 Examples : IT engineer working to create reports
  • #20 Thank you for your time today. Hope this has been helpful.