BETTER, FASTER, CHEAPER
INFRASTRUCTURE: APACHE
CLOUDSTACK + RIAK CS
John Burwell
(jburwell@basho.com | @john_burwell)

Friday, November 8, 13
Who Am I
•

Consulting Engineer @ Basho Technologies

•

Member, Apache CloudStack PMC

•

Ran operations and designed automated provisioning for hybrid
analytic/virtualization clouds

•

Led architectural design and server-side development of a SaaS
physical security platform

Friday, November 8, 13
... but I submitted a ticket.

Friday, November 8, 13
Shifting Role Of IT

From Cost Center
To Business Opportunity

Friday, November 8, 13
Shifting Role Of IT

From Cost Center
To Business Opportunity

Infrastructure Be Boring

Friday, November 8, 13
Better

•

Resilience

•

Scalability

•

Service

Friday, November 8, 13
Faster

•

Resource Provisioning

•

Systems

•

Failure recovery

Friday, November 8, 13
Cheaper

•

Hardware

•

Labor

•

Hosting

Friday, November 8, 13
Composable Infrastructure
•

Resource Pooling

•

Rapid Elasticity

•

Self Service

•

Metering/Charge Back

Friday, November 8, 13
Composable Infrastructure
•

Resource Pooling

•

Rapid Elasticity

•

Self Service

•

Metering/Charge Back

Late-binding of hardware roles

Friday, November 8, 13
Got It ... Why Go Private?

•

Leverage existing infrastructure investment

•

Customization and Control

•

Regulatory/Security Requirements

•

Cost

Friday, November 8, 13
Doesn’t vSphere/
XCP/XenServer
already do that?

Friday, November 8, 13
Friday, November 8, 13
Hopefully not ...

Friday, November 8, 13
Friday, November 8, 13
Traditional Workload
•
•

Fault tolerant

•

Management

Reliable

Specialized hardware

•

IT managed provisioning

•

Static scale out model

Friday, November 8, 13

Network

Hypervisor
Cluster

Hypervisor
Cluster

Shared
 Storage
Traditional Workload
•
•

Fault tolerant

•

Management

Reliable

Specialized hardware

•

IT managed provisioning

•

Static scale out model

Network

Hypervisor
Cluster

Hypervisor
Cluster

Shared
 Storage

Focused on resource consolidation not agility

Friday, November 8, 13
Software Is Eating Hardware
•

Moore’s Law has pulled fault tolerance up the stack

•

Commodity Hardware
•
•

•

Reduced cost
Services more workload types

Dynamic resource allocation

Friday, November 8, 13
Software Is Eating Hardware
•

Moore’s Law has pulled fault tolerance up the stack

•

Commodity Hardware
•
•

•

Reduced cost
Services more workload types

Dynamic resource allocation

Specialized Hardware Reduces Agility

Friday, November 8, 13
Cloud Workload
•

Reliable

•

Fault Tolerant

•

Self-Service provisioning

•

Control Plane - Dynamic
scale out model

Rack

Rack

App

App

Commodity Hardware

•

SDN

Friday, November 8, 13

Object
 Storage
Cloud Workload
•

Reliable

•

Fault Tolerant

•

Self-Service provisioning

•

Control Plane - Dynamic
scale out model

Rack

Rack

App

App

Commodity Hardware

•

SDN

Object
 Storage

Resource optimization increases agility

Friday, November 8, 13
Reality

Traditional
Workload

Friday, November 8, 13

Cloud
Workload
Complete Private Cloud

+

Compute Services

Friday, November 8, 13

Object Storage
Services
What Is Apache CloudStack?

§Open source software that deploys
and manages large networks of
virtual machines, as a highly
available, highly scalable
Infrastructure as a Service (IaaS)
cloud computing platform.
§CloudStack is a Top-Level Project at
the Apache Software Foundation.

Friday, November 8, 13
What Is Apache CloudStack?

§Open source software that deploys
and manages large networks of
virtual machines, as a highly
available, highly scalable
Infrastructure as a Service (IaaS)
cloud computing platform.
§CloudStack is a Top-Level Project at
the Apache Software Foundation.

Friday, November 8, 13
~300: Acknowledged production implementations
~43,000: Hosts in a single CloudStack instance
~85: Project committers

Friday, November 8, 13
Apache CloudStack Users

Friday, November 8, 13
Anatomy Of CloudStack
•

Management Server Farm
•

•

VM

Network
Host

Collection of clusters in the same failure boundary

Primary
Storage

A grouping of hosts and their associated storage

Hosts
•

•

Collection of pods, network offerings and secondary storage

Cluster
•

•

Host

Pod
•

•

VM

Zone
•

•

Management and provisioning tasks

Cluster

Secondary
Storage

Cluster

Servers onto which services will be provisioned

Primary Storage
CloudStack Pod

•
•

VM disk storage

Network
•

CloudStack Pod

Logical network associated with service offerings
Zone

•

Secondary Storage
•

Template, snapshot and ISO storage

Friday, November 8, 13
Management Server
•

Orchestrates infrastructure management

•

A single management server instance can manage 10,000 hosts

•

For production, run n +1 management servers with at least one
instance in each zone

•

Cluster MySQL

Friday, November 8, 13
Control And Data Planes
•

Automation control of
resources (e.g. hypervisors,
storage, switches, etc)

•

Small, linear bandwidth
consumption

•

Low latency

Friday, November 8, 13

•

Transfer of data between
infrastructure components
(e.g. templates, snapshots,
etc)

•

Large, variable bandwidth
consumption

•

High throughput
Control And Data Plane Scalability

•

Separate the control and data plane compute and network
traffic

•

Data transfer operations offloaded to system virtual machines

•

Separate management and storage networks

Friday, November 8, 13
Primary Vs Secondary Storage
•

Random access read/write

•

Write once/read many

•

Changes often

•

Append only

•

Zone Scoped

•

Zone or Region Scoped

•

Focused on throughput

•

Focused on reliability

Friday, November 8, 13
Primary Vs Secondary Storage
•

Random access read/write

•

Write once/read many

•

Changes often

•

Append only

•

Zone Scoped

•

Zone or Region Scoped

•

Focused on throughput

•

Focused on reliability

Secondary Storage is the lifeboat.

Friday, November 8, 13
What Is Riak CS?
•

Simple, reliable storage of large objects accessed using the S3 or Swift
API

•

Objects are stored in buckets with a unique key

•

Multi-tenant security model with per object ACLs

•

Usage Accounting and Reporting

•

Open Source (Apache2 Licensed)

Friday, November 8, 13
What Is Riak CS?
•

Simple, reliable storage of large objects accessed using the S3 or Swift
API

•

Objects are stored in buckets with a unique key

•

Multi-tenant security model with per object ACLs

•

Usage Accounting and Reporting

•

Open Source (Apache2 Licensed)

Your Own Private S3

Friday, November 8, 13
Anatomy Of Riak CS
•

S3/Swift API Proxy + Riak key-value store

•

Peer-to-Peer Design: Any node can service any read or accept
any write request

•

Maintains multiple replicas (default 3) spread across multiple
nodes

•

Ops Friendly

Friday, November 8, 13
Anatomy Of Riak CS
•

S3/Swift API Proxy + Riak key-value store

•

Peer-to-Peer Design: Any node can service any read or accept
any write request

•

Maintains multiple replicas (default 3) spread across multiple
nodes

•

Ops Friendly

Horizontally scaleable storage on
commodity hardware

Friday, November 8, 13
Riak CS Use Cases

•

Secondary Storage for VM snapshot and template data

•

Database Backup Storage

•

Serving Static Content

•

Application Log Storage

Friday, November 8, 13
Achieving Commodity Reliability

•

Create a zone per power source/backbone network connection

•

Establish independent primary storage in each zone

•

Utilize intra-datacenter block-level replication between zones

•

Distribute Riak CS nodes evenly across zones

Friday, November 8, 13
The Complete Private Cloud
Traditional

Better, Faster, Cheaper Infrastructure: Apache CloudStack and Riak CS

  • 1.
    BETTER, FASTER, CHEAPER INFRASTRUCTURE:APACHE CLOUDSTACK + RIAK CS John Burwell (jburwell@basho.com | @john_burwell) Friday, November 8, 13
  • 2.
    Who Am I • ConsultingEngineer @ Basho Technologies • Member, Apache CloudStack PMC • Ran operations and designed automated provisioning for hybrid analytic/virtualization clouds • Led architectural design and server-side development of a SaaS physical security platform Friday, November 8, 13
  • 3.
    ... but Isubmitted a ticket. Friday, November 8, 13
  • 4.
    Shifting Role OfIT From Cost Center To Business Opportunity Friday, November 8, 13
  • 5.
    Shifting Role OfIT From Cost Center To Business Opportunity Infrastructure Be Boring Friday, November 8, 13
  • 6.
  • 7.
  • 8.
  • 9.
    Composable Infrastructure • Resource Pooling • RapidElasticity • Self Service • Metering/Charge Back Friday, November 8, 13
  • 10.
    Composable Infrastructure • Resource Pooling • RapidElasticity • Self Service • Metering/Charge Back Late-binding of hardware roles Friday, November 8, 13
  • 11.
    Got It ...Why Go Private? • Leverage existing infrastructure investment • Customization and Control • Regulatory/Security Requirements • Cost Friday, November 8, 13
  • 12.
    Doesn’t vSphere/ XCP/XenServer already dothat? Friday, November 8, 13
  • 13.
  • 14.
  • 15.
  • 16.
    Traditional Workload • • Fault tolerant • Management Reliable Specializedhardware • IT managed provisioning • Static scale out model Friday, November 8, 13 Network Hypervisor Cluster Hypervisor Cluster Shared
  • 17.
  • 18.
    Traditional Workload • • Fault tolerant • Management Reliable Specializedhardware • IT managed provisioning • Static scale out model Network Hypervisor Cluster Hypervisor Cluster Shared
  • 19.
     Storage Focused on resourceconsolidation not agility Friday, November 8, 13
  • 20.
    Software Is EatingHardware • Moore’s Law has pulled fault tolerance up the stack • Commodity Hardware • • • Reduced cost Services more workload types Dynamic resource allocation Friday, November 8, 13
  • 21.
    Software Is EatingHardware • Moore’s Law has pulled fault tolerance up the stack • Commodity Hardware • • • Reduced cost Services more workload types Dynamic resource allocation Specialized Hardware Reduces Agility Friday, November 8, 13
  • 22.
    Cloud Workload • Reliable • Fault Tolerant • Self-Serviceprovisioning • Control Plane - Dynamic scale out model Rack Rack App App Commodity Hardware • SDN Friday, November 8, 13 Object
  • 23.
  • 24.
    Cloud Workload • Reliable • Fault Tolerant • Self-Serviceprovisioning • Control Plane - Dynamic scale out model Rack Rack App App Commodity Hardware • SDN Object
  • 25.
     Storage Resource optimization increasesagility Friday, November 8, 13
  • 26.
  • 27.
    Complete Private Cloud + ComputeServices Friday, November 8, 13 Object Storage Services
  • 28.
    What Is ApacheCloudStack? §Open source software that deploys and manages large networks of virtual machines, as a highly available, highly scalable Infrastructure as a Service (IaaS) cloud computing platform. §CloudStack is a Top-Level Project at the Apache Software Foundation. Friday, November 8, 13
  • 29.
    What Is ApacheCloudStack? §Open source software that deploys and manages large networks of virtual machines, as a highly available, highly scalable Infrastructure as a Service (IaaS) cloud computing platform. §CloudStack is a Top-Level Project at the Apache Software Foundation. Friday, November 8, 13
  • 30.
    ~300: Acknowledged productionimplementations ~43,000: Hosts in a single CloudStack instance ~85: Project committers Friday, November 8, 13
  • 31.
  • 32.
    Anatomy Of CloudStack • ManagementServer Farm • • VM Network Host Collection of clusters in the same failure boundary Primary Storage A grouping of hosts and their associated storage Hosts • • Collection of pods, network offerings and secondary storage Cluster • • Host Pod • • VM Zone • • Management and provisioning tasks Cluster Secondary Storage Cluster Servers onto which services will be provisioned Primary Storage CloudStack Pod • • VM disk storage Network • CloudStack Pod Logical network associated with service offerings Zone • Secondary Storage • Template, snapshot and ISO storage Friday, November 8, 13
  • 33.
    Management Server • Orchestrates infrastructuremanagement • A single management server instance can manage 10,000 hosts • For production, run n +1 management servers with at least one instance in each zone • Cluster MySQL Friday, November 8, 13
  • 34.
    Control And DataPlanes • Automation control of resources (e.g. hypervisors, storage, switches, etc) • Small, linear bandwidth consumption • Low latency Friday, November 8, 13 • Transfer of data between infrastructure components (e.g. templates, snapshots, etc) • Large, variable bandwidth consumption • High throughput
  • 35.
    Control And DataPlane Scalability • Separate the control and data plane compute and network traffic • Data transfer operations offloaded to system virtual machines • Separate management and storage networks Friday, November 8, 13
  • 36.
    Primary Vs SecondaryStorage • Random access read/write • Write once/read many • Changes often • Append only • Zone Scoped • Zone or Region Scoped • Focused on throughput • Focused on reliability Friday, November 8, 13
  • 37.
    Primary Vs SecondaryStorage • Random access read/write • Write once/read many • Changes often • Append only • Zone Scoped • Zone or Region Scoped • Focused on throughput • Focused on reliability Secondary Storage is the lifeboat. Friday, November 8, 13
  • 38.
    What Is RiakCS? • Simple, reliable storage of large objects accessed using the S3 or Swift API • Objects are stored in buckets with a unique key • Multi-tenant security model with per object ACLs • Usage Accounting and Reporting • Open Source (Apache2 Licensed) Friday, November 8, 13
  • 39.
    What Is RiakCS? • Simple, reliable storage of large objects accessed using the S3 or Swift API • Objects are stored in buckets with a unique key • Multi-tenant security model with per object ACLs • Usage Accounting and Reporting • Open Source (Apache2 Licensed) Your Own Private S3 Friday, November 8, 13
  • 40.
    Anatomy Of RiakCS • S3/Swift API Proxy + Riak key-value store • Peer-to-Peer Design: Any node can service any read or accept any write request • Maintains multiple replicas (default 3) spread across multiple nodes • Ops Friendly Friday, November 8, 13
  • 41.
    Anatomy Of RiakCS • S3/Swift API Proxy + Riak key-value store • Peer-to-Peer Design: Any node can service any read or accept any write request • Maintains multiple replicas (default 3) spread across multiple nodes • Ops Friendly Horizontally scaleable storage on commodity hardware Friday, November 8, 13
  • 42.
    Riak CS UseCases • Secondary Storage for VM snapshot and template data • Database Backup Storage • Serving Static Content • Application Log Storage Friday, November 8, 13
  • 43.
    Achieving Commodity Reliability • Createa zone per power source/backbone network connection • Establish independent primary storage in each zone • Utilize intra-datacenter block-level replication between zones • Distribute Riak CS nodes evenly across zones Friday, November 8, 13
  • 44.
    The Complete PrivateCloud Traditional