1
HYPER-CONVERGED INFRASTRUCTURE STRATEGIES MYTH BUSTING & THE FUTURE OF WEB SCALE IT 2
ROADMAP INFORMATION DISCLAIMER EMC makes no representation and undertakes no obligations with regard to product planning information, anticipated product characteristics, performance specifications, or anticipated release dates (collectively, Roadmap Information ). Roadmap Information is provided by EMC as an accommodation to the recipient solely for purposes of discussion and without intending to be bound thereby. Roadmap information is EMC Restricted Confidential and is provided under the terms, conditions and restrictions defined in the EMC Non- Disclosure Agreement in place with your organization.
TOPICS Hype Road to Hyper-Convergence Myths Design Considerations, Benefits, and Trade-Offs EMC ETD Hyper-Converged Product Strategy
HYPE 5
Hyper-convergence is the confluence of 3 industry trends.
1 Software-Defined Everything 2 Commodity Hardware 3 Scale-out Design
Trigger: The emergence of software-defined storage.
The promise is lower capital costs, lower operating costs, and increased operational agility.
ROAD TO HYPER- CONVERGENCE 10
Move to scale-out design was triggered by the end of Moore s Law.
Software architectures moved to a design center of utilizing many CPU cores vs. waiting for faster ones.
ROAD TO HYPER-CONVERGENCE Scale Out vs Scale Up A B A B A B C N... Vertical Scaling Make boxes bigger (usually an HA pair or master election system) Horizontal Scaling Make more boxes
Such scale-out techniques were applied to both application and infrastructure software design.
ROAD TO HYPER- CONVERGENCE Server Server Server Server Server Server Storage Array IP SAN Server Server Server Storage Array Server Server Server
ROAD TO HYPER-CONVERGENCE Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Storage Array IP SAN Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Storage Array Managed as a scale-out compute farm Apps deployed in VMs or Containers
ROAD TO HYPER- CONVERGENCE Managed as a scale-out compute farm Managed as a scale-out storage farm Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 IP Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Apps deployed in VMs or Containers Scale-out SDS deployed in VMs or Containers
AND NOW, THE $64K QUESTION Are two commodity farms needed? Or can storage and compute run on the same one?
ROAD TO HYPER-CONVERGENCE Apps & SDS deployed in VMs or Containers Commodity x86 Commodity x86 Commodity x86 Commodity x86 Scale-out Compute Management IP Scale-out Storage Management Commodity x86 Commodity x86 Commodity x86 Commodity x86
ROAD TO HYPER-CONVERGENCE Apps & SDS deployed in VMs or Containers App Only Nodes Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Scale-out Compute Management Mixed Nodes IP Mixed Nodes Scale-out Storage Management Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Commodity x86 Storage Only Nodes
MYTHS 21
MYTHS Hyper-convergence is only for small scale.
MYTHS You can t scale storage and compute independently.
MYTHS Workloads must be virtualized.
MYTHS All hardware has to be the same form factor.
MYTHS I have to buy hardware and software from the same vendor.
MYTHS Hyper-convergence is the answer to everything.
DESIGN CONSIDERATIONS, BENEFITS, TRADE-OFFS 28
SCENARIO Key Metrics: $/VM Key Metrics: Response Time, $/Outage Key Metrics: $/GB, Throughput Key Metrics: $/GB, $/Outage, Geo General Compute Farm VMs for web servers, mobile app servers, and other stateless applications (No)SQL Database Farm VMs hosting semi-structured (MongoDB, Cassandra) and structured (MySQL) databases Hadoop Analytics Farm Bare Metal batch (map/reduce) and real-time (spark streaming) analytics Content Repository Bare Metal Object Storage for video, images, and other unstructured content Hyper-Converged Infrastructure
DESIGN CONSIDERATIONS: HARDWARE HOMOGENEOUS OR HETEROGENEOUS? $/VM Response Time, $/Outage $/GB, Throughput $/GB, $/Outage, Geo General Compute Farm (No)SQL Database Farm Hadoop Analytics Farm Content Repository 1 Dense CPU 2 High I/O x24 x24 x24 x24 x24 x24 x24 x24 x24 3 Workload Optimized x24 x24 x24 x24 x24 x60 x60
DESIGN CONSIDERATIONS: SCALING SCALE STORAGE AND COMPUTE LINEARLY OR INDEPENDENTLY? $/VM Response Time, $/Outage $/GB, Throughput $/GB, $/Outage, Geo General Compute Farm (No)SQL Database Farm Hadoop Analytics Farm Content Repository Stateless Apps Hyper visor SDS (Block) Hyper visor Databases SDS (Block) Hadoop/Spark Bare Metal OS SDS (HDFS) SDS (Object, HDFS) Linear Scaling Stateless apps + low capacity VM disks x24 x24 Linear Scaling DBs + high capacity VM disks x24 x24 Linear Scaling Hadoop + cluster local HDFS storage x60 x60 Independent Scaling Shared storage
DESIGN CONSIDERATIONS: SDS STACK ONE STACK VS. MULTIPLE STACKS? $/VM Response Time, $/Outage $/GB, Throughput $/GB, $/Outage, Geo General Compute Farm (No)SQL Database Farm Hadoop Analytics Farm Content Repository Stateless Apps Hyper visor SDS (Block) Hyper visor Databases SDS (Block) Hadoop/Spark Bare Metal OS SDS (HDFS) SDS (Object, HDFS) Low Latency, Random I/O Conflicting Design Centers High Throughput, Streaming I/O = Design Center for ScaleIO = Design Center for ECS Software vs. Single SDS Stack for Both (compromise in one or the other)
DESIGN CONSIDERATIONS: OPS MODEL INTEGRATED DEV OPS VS. TRADITIONAL COMPUTE AND STORAGE OPS TEAMS $/VM General Compute Farm Response Time, $/Outage (No)SQL Database Farm $/GB, Throughput Hadoop Analytics Farm $/GB, $/Outage, Geo Content Repository 1 2 Stateless Apps Hyper visor SDS (Block) Hyper visor Databases SDS (Block) Hadoop/Spark Bare Metal OS SDS (HDFS) SDS (Object, HDFS) Dev Ops Unit Dev Ops Unit Dev Ops Unit Dev Ops Unit Dev Ops Unit Dev Ops Unit Dev Ops Unit 3 Dev Ops Unit Dev Ops Unit
EMC ETD HYPER-CONVERGED PRODUCT STRATEGY 34
PRINCIPLES 1 World Class SDS 2 Flexibility and Choice 3 @Scale
EMC ETD HYPER-CONVERGED PRODUCT STRATEGY World-class SDS Solutions for block and unstructured storage: ScaleIO and ECS Software.
EMC ETD HYPER-CONVERGED PRODUCT STRATEGY Building Block CI Solutions bundling ScaleIO, CI M&O, and EMC commodity hardware: Project Buzz.
EMC ETD HYPER-CONVERGED PRODUCT STRATEGY Open Cloud CI Solutions bundling open source compute OpenStack and Hadoop, to start with ScaleIO, ECS Software, CI M&O, and commodity hardware: Project Caspian.
SUMMARY 39
HYPER-CONVERGENCE SUMMARY 1 SDS + Commodity + Scale-out 2 SDS is enabler: place bets carefully 3 Ops model must align 4 Choice is key: not one-size-fits-all