+
Robert Novak, Consulting SE at Cisco
Bill Peterson, Sr. Director, Industry Solutions at MapR
Bringing Structure, Scalability, and
Services to Cloud-Scale Storage
Storage and analysis of data, whether logs or
images or sound or video, has long been a
concern in computing.
NIST/NBS image from 1960
1. DAS vs NAS vs SAN vs THE CLOUD
2. MapR-XD storage platform
and Converge-X Data Fabric
3. Cisco UCS for MapR-XD deployments
4. Where MapR-XD fits in
the modern enterprise
5. Don’t try this at home:
Where MapR-XD doesn’t fit
6. Where do we go from here?
What are we talking about today?
DAS vs NAS
vs SAN vs the CLOUD
Storage models in the traditional datacenter
• DAS – Direct Attach Storage (everyone uses it, block and file)
• Hadoop depends on this model, as do most personal devices
• NAS – Network Attach Storage (file storage over a LAN)
• Broadly shared files and folders, central management
• SAN – Storage Area Network (switched block storage)
• Higher performance, separate network, highest cost (usually)
• Cloud – Object, block, and file over a larger network distance
• The effect of any of the above, but on Someone Else’s Computer™
Storage evolution since the 1990s
• DAS – Managed per server, crash cart city, difficult to reallocate
performance and capacity
• NAS – Shared performance for legacy apps, harder to guarantee
performance
• SAN – Expensive, with special hardware required on clients
• Cloud – Capacity, performance, and price are quickly expandable,
and can be challenging to use remotely
Storage limitations since the 1990s
• Scaling NAS and SAN can require,
well, you know…
• Backup and restore, disaster recovery
tend to be bolt-ons/afterthoughts
• Need millions of files, or lots of tiny
files, or a mix of flash and spinning
disk? Good luck!
But wait, there’s more…
What if…
Your scalable storage platform
provided industry-standard access
methods like NFS
Existing applications like SAP HANA
could plug right in with no recoding
New applications, including the Hadoop
ecosystem, could live side by side with
traditional applications in a single
management framework
You could scale storage performance
and storage capacity independently
With Cisco UCS
(Unified Computing System)
and the MapR-XD solution…
Your infrastructure
can do just that.
Amazing but true.
(Tell me more?)
The MapR Advantage
MapR-XD storage platform for enterprise grade
scalability, performance, and reliability
© 2017 MapR TechnologiesMapR Confidential 10
Industry Context For MapR Opportunity
Modernize Data
Storage and
Management
Multi-
Infrastructure
(on prem, cloud and
edge)
Massive Scale
& Heterogeneity
Host Legacy &
New
applications
MapR
opportunity
© 2017 MapR TechnologiesMapR Confidential 11
New Use Cases & Workloads Are Changing “Storage”
Scientific
Apps
Machine
Data
Analytics +
Log
Image/video
Processing Build
Test
Containers
Global
Repository
ERP
CRM
Cold Data Backup
Home
Dir
OLTP
VMs
Perf
Capacity
Delivered in a cloud-scale architecture with
commodity economics across public and private
clouds.
Workloads
▪ Legacy Operational apps
▪ Media & Entertainment
▪ Image Processing
▪ Analytics & Log Processing
▪ Container based apps
Capabilities
▪ Global Namespace
▪ Multi-temperature store
▪ Extreme Scale & Reliability
▪ Guaranteed Performance
▪ Agile Platform for
heterogeneous data types
▪ Integrated Analytics
© 2017 MapR TechnologiesMapR Confidential 12
HDFSS3POSIX Geo-Replication
Self Healing
Assume Failures are
Common
Many Small Sites
A Few Big Sites
Leverages Cloud
Globally Protected
Globally Accessible
Globally Managed
NFS Data Protection
What customers are telling us?
Need Scalable Storage Platform Around the globe, Around the clock
© 2017 MapR TechnologiesMapR Confidential 13
Technology Disruption Trends
FLASH
Harness Flash
disruption
SOFTWARE
DEFINED
Decouple
Software from
Hardware
CLOUD
Bridge gap
between on-prem
& Cloud with Cost
effective Services
SCALE
Scale
Capacity,
Performance
© 2017 MapR TechnologiesMapR Confidential 14
Multi Cloud Scale out to 1000s of Nodes Enterprise Grade Simple & Efficient to Operate
MapR-XD Supports A Broad Range Of Data Types
Files & Folders Containers Custom Apps Hadoop/Spark Apps
Heterogeneous
Nodes
300
PB
10+
EB
MapR-XD
Powered by the Converge-X Data Fabric
Management
EDg
Edge Clusters Multi-cloud
MapR
CORE
EdgeEdge
Edge
Deploying MapR-XD on
Cisco UCS M5 servers
Cisco UCS excels for scalable deployments
• Single point of control for servers,
network, storage
• Policy-driven configuration and
firmware management
• High performance networking at
the core
• Flexible compute, network, and
storage options
• Open XML API for automation
Cisco UCS Baseline Elements
• Fabric Interconnects (6332)
provide management and 10/40
Gigabit Ethernet connectivity
• C240 M5S servers provide 2-
socket 24-SFF-drive platform with
up to 3TB RAM each
• S3260 servers provide 1-2
2-socket server nodes per 4U
chassis, with up to 56 LFF drives
• Models can be mixed per domain
More details on Cisco UCS 6300 Series Fabric Interconnects,
Cisco UCS C240 M5 servers, and Cisco UCS S3260 servers
available on Cisco.com.
Let's look at some sample configurations
These are just examples to illustrate the platform.
Cisco and MapR are validating and optimizing our joint reference architectures
for MapR-XD on the new Cisco UCS M5 rackmount servers.
Stay tuned for updated architectures at cisco.com/go/bigdata
Performance-oriented configurations (C240)
• Minimum 5 servers per cluster
• 6300-series Fabric Interconnects
• C240 M5S 24-drive SFF model
• Dual Intel Xeon Scalable CPU
• 128GB minimum DDR4 RAM
• 8 servers provide 345TB raw
(with 1.8TB 10k SAS drives)
• 16 servers provide 690TB raw
Capacity Optimized Configurations (S3260)
• Minimum 5 servers per cluster
• 6300-series Fabric Interconnects
• S3260 Storage-Optimized Server,
2 server nodes per chassis
• Dual Intel Xeon Scalable CPU
• 3 Chassis (6 servers) provide
1.6PB raw (with 10TB drives)
• 6 Chassis (12 servers) provide
3.2PB rawMore details on Cisco UCS S3260 servers
available on Cisco.com, including different
drive type and size options.
Hybrid Multi-Temperature Configuration
• 6300-series Fabric Interconnects
with 40 Gigabit Ethernet
• 8-server C240 set (345TB raw)
for most active data
• 6-server S3260 set (3-chassis,
3.2PB raw)
for higher volume data and/or
long-term archiving
More details on Cisco UCS S3260 servers available on
Cisco.com, including different drive type and size options.
Applying MapR-XD to the
modern data ecosystem
The perfect fit for your cloud-scale data needs
© 2017 MapR TechnologiesMapR Confidential 23
MapR-XD Customer Value Proposition
Global Namespace
Eliminate
Silos
CoreEdge Cloud
Web
Apps
File
Apps
Any Data,
Any Workload,
One Platform
Container
Apps
S3NFS HDFS POSIX
Data Aware
MapR-XD
Deploy, Execute
Anywhere
Public
Cloud
Commodity
Servers
Reduce
TCO
MapR-XD
Cloud
Scale
HOT
WARM
COLD
A View of How Data is Managed
• Data Splits into 2 tiers
– High Performance
• Delivered using flash
• Latency under ms, millions of IOPs,
multi-GBs of throughput
– Long term data
• Retained forever
• Delivered by disk / public-cloud
• Massively scalable and geo distributed
• Moves rarely accessed data to cold tier
• Simplicity & Efficiency @ scale is key
– Machines will create/analyze/move data around.
– DR/BR/Archive storage is merging
– Hybrid deployment model is here to stay
– Available 24x7 (non disruptive operation)
Public CloudNew York. London
Operational
Apps
Analytical
Apps
Modern
Web2.0 Apps
Object*
Frequently Accessed Data
HDFS File
Infrequently Accessed Data
Rarely Access Data
Containers
* Roadmap
© 2017 MapR TechnologiesMapR Confidential 25
Multi-Cluster
Dashboard
•Health indicators
•Categorized alerts
•Activity/utilization
Cloud Platform
Multi-PB, Multi-Tier
• Competed against IBM Spectrum Scale
• Determined IO was 3-4X faster than legacy
• Requirements: reliable, durable, automatic failover,
replication, snapshots, mirroring, per-tenant security,
logical encryption
© 2017 MapR TechnologiesMapR Confidential 27
Connected Car
Use Cases
• Advanced Driver
Assistance Systems
(ADAS)
• Computer Aided Driving
• Vehicle Healthcare
• Fleet Management
By 2020, more than 250
million vehicles will be
connected globally
© 2017 MapR TechnologiesMapR Confidential 28
Why Samsung bet on MapR for “Real Time Apps”
Voice Command Processing
(AI,GPU)
MapR-XD
Scalable server platform
✓ Enterprise
Grade
✓ Trillions of
Files
✓ Reduce
Cost
Don’t try this at home, folks
Where MapR-XD won’t be your first choice (yet)
• MapR-XD provides an excellent
platform for many storage models
• One tool never fits all needs, though
• Just because you can, doesn’t
mean you should
There is no Universal Business Adapter
• Plug-and-play NAS (replacement)
• Native block and object storage
• Appliance-style deployment
• IOPs-focused platform
• Orbital death ray automation
Don’t try these at home, folks
• File-based storage platform with rapid
growth
• Vast numbers of files, of any size
• Seamless multi-site replication,
whether for geo-redundancy,
disaster recovery, or compliance
• Throughput-optimized platform,
especially for large/sequential data
• Analytics platform for orbital death rays
Try these on for size
Where do we go from here?
Resources and next steps
for your Converged Data Platform journey
Q&A
ENGAGE WITH US
Contact us at:
855-NOW-MAPR
Additional Resources
• Learn more about MapR-XD at: https://mapr.com/products/mapr-xd/
• See the latest developments with Cisco UCS at: cisco.com/go/ucs
• MapR & Cisco Make IT Better – previous webinar series episodes
1. It's No Use Going Back to Yesterday's Storage Platform for Tomorrow's Applications (Available on mapr.com on-demand)
2. Cisco & MapR bring 3 Superpowers to SAP HANA Deployments (available on mapr.com on-demand)
• Follow us at: @gallifreyan, @mapr, @Cisco, and @CiscoDC on Twitter
• Meet us at MapR Convergence, & Strata+Hadoop World
• For the MapR latest books, whitepapers, webinars, etc. visit: https://mapr.com/resources/
• For Cisco’s Big Data and Analytics solutions, visit cisco.com/go/bigdata
Thank you for
joining us today.

Bringing Structure, Scalability, and Services to Cloud-Scale Storage

  • 1.
    + Robert Novak, ConsultingSE at Cisco Bill Peterson, Sr. Director, Industry Solutions at MapR Bringing Structure, Scalability, and Services to Cloud-Scale Storage
  • 2.
    Storage and analysisof data, whether logs or images or sound or video, has long been a concern in computing. NIST/NBS image from 1960 1. DAS vs NAS vs SAN vs THE CLOUD 2. MapR-XD storage platform and Converge-X Data Fabric 3. Cisco UCS for MapR-XD deployments 4. Where MapR-XD fits in the modern enterprise 5. Don’t try this at home: Where MapR-XD doesn’t fit 6. Where do we go from here? What are we talking about today?
  • 3.
    DAS vs NAS vsSAN vs the CLOUD Storage models in the traditional datacenter
  • 4.
    • DAS –Direct Attach Storage (everyone uses it, block and file) • Hadoop depends on this model, as do most personal devices • NAS – Network Attach Storage (file storage over a LAN) • Broadly shared files and folders, central management • SAN – Storage Area Network (switched block storage) • Higher performance, separate network, highest cost (usually) • Cloud – Object, block, and file over a larger network distance • The effect of any of the above, but on Someone Else’s Computer™ Storage evolution since the 1990s
  • 5.
    • DAS –Managed per server, crash cart city, difficult to reallocate performance and capacity • NAS – Shared performance for legacy apps, harder to guarantee performance • SAN – Expensive, with special hardware required on clients • Cloud – Capacity, performance, and price are quickly expandable, and can be challenging to use remotely Storage limitations since the 1990s
  • 6.
    • Scaling NASand SAN can require, well, you know… • Backup and restore, disaster recovery tend to be bolt-ons/afterthoughts • Need millions of files, or lots of tiny files, or a mix of flash and spinning disk? Good luck! But wait, there’s more…
  • 7.
    What if… Your scalablestorage platform provided industry-standard access methods like NFS Existing applications like SAP HANA could plug right in with no recoding New applications, including the Hadoop ecosystem, could live side by side with traditional applications in a single management framework You could scale storage performance and storage capacity independently
  • 8.
    With Cisco UCS (UnifiedComputing System) and the MapR-XD solution… Your infrastructure can do just that. Amazing but true. (Tell me more?)
  • 9.
    The MapR Advantage MapR-XDstorage platform for enterprise grade scalability, performance, and reliability
  • 10.
    © 2017 MapRTechnologiesMapR Confidential 10 Industry Context For MapR Opportunity Modernize Data Storage and Management Multi- Infrastructure (on prem, cloud and edge) Massive Scale & Heterogeneity Host Legacy & New applications MapR opportunity
  • 11.
    © 2017 MapRTechnologiesMapR Confidential 11 New Use Cases & Workloads Are Changing “Storage” Scientific Apps Machine Data Analytics + Log Image/video Processing Build Test Containers Global Repository ERP CRM Cold Data Backup Home Dir OLTP VMs Perf Capacity Delivered in a cloud-scale architecture with commodity economics across public and private clouds. Workloads ▪ Legacy Operational apps ▪ Media & Entertainment ▪ Image Processing ▪ Analytics & Log Processing ▪ Container based apps Capabilities ▪ Global Namespace ▪ Multi-temperature store ▪ Extreme Scale & Reliability ▪ Guaranteed Performance ▪ Agile Platform for heterogeneous data types ▪ Integrated Analytics
  • 12.
    © 2017 MapRTechnologiesMapR Confidential 12 HDFSS3POSIX Geo-Replication Self Healing Assume Failures are Common Many Small Sites A Few Big Sites Leverages Cloud Globally Protected Globally Accessible Globally Managed NFS Data Protection What customers are telling us? Need Scalable Storage Platform Around the globe, Around the clock
  • 13.
    © 2017 MapRTechnologiesMapR Confidential 13 Technology Disruption Trends FLASH Harness Flash disruption SOFTWARE DEFINED Decouple Software from Hardware CLOUD Bridge gap between on-prem & Cloud with Cost effective Services SCALE Scale Capacity, Performance
  • 14.
    © 2017 MapRTechnologiesMapR Confidential 14 Multi Cloud Scale out to 1000s of Nodes Enterprise Grade Simple & Efficient to Operate MapR-XD Supports A Broad Range Of Data Types Files & Folders Containers Custom Apps Hadoop/Spark Apps Heterogeneous Nodes 300 PB 10+ EB MapR-XD Powered by the Converge-X Data Fabric Management EDg Edge Clusters Multi-cloud MapR CORE EdgeEdge Edge
  • 15.
  • 16.
    Cisco UCS excelsfor scalable deployments • Single point of control for servers, network, storage • Policy-driven configuration and firmware management • High performance networking at the core • Flexible compute, network, and storage options • Open XML API for automation
  • 17.
    Cisco UCS BaselineElements • Fabric Interconnects (6332) provide management and 10/40 Gigabit Ethernet connectivity • C240 M5S servers provide 2- socket 24-SFF-drive platform with up to 3TB RAM each • S3260 servers provide 1-2 2-socket server nodes per 4U chassis, with up to 56 LFF drives • Models can be mixed per domain More details on Cisco UCS 6300 Series Fabric Interconnects, Cisco UCS C240 M5 servers, and Cisco UCS S3260 servers available on Cisco.com.
  • 18.
    Let's look atsome sample configurations These are just examples to illustrate the platform. Cisco and MapR are validating and optimizing our joint reference architectures for MapR-XD on the new Cisco UCS M5 rackmount servers. Stay tuned for updated architectures at cisco.com/go/bigdata
  • 19.
    Performance-oriented configurations (C240) •Minimum 5 servers per cluster • 6300-series Fabric Interconnects • C240 M5S 24-drive SFF model • Dual Intel Xeon Scalable CPU • 128GB minimum DDR4 RAM • 8 servers provide 345TB raw (with 1.8TB 10k SAS drives) • 16 servers provide 690TB raw
  • 20.
    Capacity Optimized Configurations(S3260) • Minimum 5 servers per cluster • 6300-series Fabric Interconnects • S3260 Storage-Optimized Server, 2 server nodes per chassis • Dual Intel Xeon Scalable CPU • 3 Chassis (6 servers) provide 1.6PB raw (with 10TB drives) • 6 Chassis (12 servers) provide 3.2PB rawMore details on Cisco UCS S3260 servers available on Cisco.com, including different drive type and size options.
  • 21.
    Hybrid Multi-Temperature Configuration •6300-series Fabric Interconnects with 40 Gigabit Ethernet • 8-server C240 set (345TB raw) for most active data • 6-server S3260 set (3-chassis, 3.2PB raw) for higher volume data and/or long-term archiving More details on Cisco UCS S3260 servers available on Cisco.com, including different drive type and size options.
  • 22.
    Applying MapR-XD tothe modern data ecosystem The perfect fit for your cloud-scale data needs
  • 23.
    © 2017 MapRTechnologiesMapR Confidential 23 MapR-XD Customer Value Proposition Global Namespace Eliminate Silos CoreEdge Cloud Web Apps File Apps Any Data, Any Workload, One Platform Container Apps S3NFS HDFS POSIX Data Aware MapR-XD Deploy, Execute Anywhere Public Cloud Commodity Servers Reduce TCO MapR-XD Cloud Scale HOT WARM COLD
  • 24.
    A View ofHow Data is Managed • Data Splits into 2 tiers – High Performance • Delivered using flash • Latency under ms, millions of IOPs, multi-GBs of throughput – Long term data • Retained forever • Delivered by disk / public-cloud • Massively scalable and geo distributed • Moves rarely accessed data to cold tier • Simplicity & Efficiency @ scale is key – Machines will create/analyze/move data around. – DR/BR/Archive storage is merging – Hybrid deployment model is here to stay – Available 24x7 (non disruptive operation) Public CloudNew York. London Operational Apps Analytical Apps Modern Web2.0 Apps Object* Frequently Accessed Data HDFS File Infrequently Accessed Data Rarely Access Data Containers * Roadmap
  • 25.
    © 2017 MapRTechnologiesMapR Confidential 25 Multi-Cluster Dashboard •Health indicators •Categorized alerts •Activity/utilization
  • 26.
    Cloud Platform Multi-PB, Multi-Tier •Competed against IBM Spectrum Scale • Determined IO was 3-4X faster than legacy • Requirements: reliable, durable, automatic failover, replication, snapshots, mirroring, per-tenant security, logical encryption
  • 27.
    © 2017 MapRTechnologiesMapR Confidential 27 Connected Car Use Cases • Advanced Driver Assistance Systems (ADAS) • Computer Aided Driving • Vehicle Healthcare • Fleet Management By 2020, more than 250 million vehicles will be connected globally
  • 28.
    © 2017 MapRTechnologiesMapR Confidential 28 Why Samsung bet on MapR for “Real Time Apps” Voice Command Processing (AI,GPU) MapR-XD Scalable server platform ✓ Enterprise Grade ✓ Trillions of Files ✓ Reduce Cost
  • 29.
    Don’t try thisat home, folks Where MapR-XD won’t be your first choice (yet)
  • 30.
    • MapR-XD providesan excellent platform for many storage models • One tool never fits all needs, though • Just because you can, doesn’t mean you should There is no Universal Business Adapter
  • 31.
    • Plug-and-play NAS(replacement) • Native block and object storage • Appliance-style deployment • IOPs-focused platform • Orbital death ray automation Don’t try these at home, folks
  • 32.
    • File-based storageplatform with rapid growth • Vast numbers of files, of any size • Seamless multi-site replication, whether for geo-redundancy, disaster recovery, or compliance • Throughput-optimized platform, especially for large/sequential data • Analytics platform for orbital death rays Try these on for size
  • 33.
    Where do wego from here? Resources and next steps for your Converged Data Platform journey
  • 34.
    Q&A ENGAGE WITH US Contactus at: 855-NOW-MAPR
  • 35.
    Additional Resources • Learnmore about MapR-XD at: https://mapr.com/products/mapr-xd/ • See the latest developments with Cisco UCS at: cisco.com/go/ucs • MapR & Cisco Make IT Better – previous webinar series episodes 1. It's No Use Going Back to Yesterday's Storage Platform for Tomorrow's Applications (Available on mapr.com on-demand) 2. Cisco & MapR bring 3 Superpowers to SAP HANA Deployments (available on mapr.com on-demand) • Follow us at: @gallifreyan, @mapr, @Cisco, and @CiscoDC on Twitter • Meet us at MapR Convergence, & Strata+Hadoop World • For the MapR latest books, whitepapers, webinars, etc. visit: https://mapr.com/resources/ • For Cisco’s Big Data and Analytics solutions, visit cisco.com/go/bigdata
  • 36.