Storage Virtualization StrategiesStorage Virtualization Strategies

Arun Taneja

Founder, Taneja Group

September 25, 2006

AgendaAgenda

• Next Generation Data Center & Virtualization

• Storage Virtualization Approaches, Architecture, & Players

• Other Types of Virtualization & their Impact on Storage

• Recommendations to End Users

Virtual Server Virtual Server Virtual Server Virtual Server

• “Pools” of commonly grouped physical resources

• Dynamic allocations based on application level grouping and usage policies

• Interconnected and controlled through an intelligent interconnect fabric

Server Processing I/O StorageApplications

Intelligent Fabric

Stand-By

Resource Pool

Compute Networking and Storage VirtualizationCompute Networking and Storage Virtualization

Virtualized Data CenterVirtualized Data Center

• Four main types of virtualization technologies emerging:

– Server Virtualization

• Virtualizes the physical CPU, Memory, I/O of servers

– I/O Virtualization

• Virtualizes the physical network topology and mappings between

servers and storage

– File Virtualization

• Virtualizes files and namespaces across file servers

– Storage Virtualization

• Virtualizes physical block storage devices

The State of Data StorageThe State of Data StorageA Snapshot of “Where We Are” Today:

• Data Tsunami Continues• Data Scattered across data centers and remote

locations• Stringent Compliance And Regulatory Environment• Data Protection is at a break point• Can’t add IT staff indiscriminately to keep up with

capacity growth• Storage management tools are archaic• Storage tiers still infancy. Data is not mobile.

Typical SAN TodayTypical SAN Today

Ultimate Vision of SAN is Incomplete

SANSAN

V1V1

V2V2V3V3

V4V4 V5V5

Server 1Server 1

V1V1

V2V2 V3V3 V4V4 V5V5

Server 2Server 2 Server 3Server 3 Server 4Server 4

Why Storage Virtualization is En Vogue Again…Why Storage Virtualization is En Vogue Again…

• Heterogeneous Storage is a Reality

• Data Migration, Data Consolidation are Disruptive Processes Today

• V has Passed the “Hype” and “Low Point” Stages and into “Reality” Stage

• Big guys have Ratified the Concept (IBM, EMC, HDS)

What is Storage Virtualization?What is Storage Virtualization?

• Separation of Physical and Logical

• Enables One Common Set of Applications

• Physical can be Heterogeneous or Homogeneous

• Underwear

RAID equals elemental storage virtualization

Heterogeneous & Homogeneous Heterogeneous & Homogeneous VirtualizationVirtualization

EMCEMC

Virtualization LayerVirtualization Layer

Common Storage ApplicationsCommon Storage Applications

EMCEMC HDSHDS JBODJBOD

What Does Storage Virtualization Enable?What Does Storage Virtualization Enable?

• Consistent Interface for All Storage Applications

• Non-disruptive LUN Creation, Expansion, Deletion

• Non-disruptive Data Consolidation, Data Migration

• Non-disruptive RAID Level Changes

• Tiering of Storage Using Hetero Storage

• Replication Across Hetero Storage

• Backup/Restore, Archival, Thin Provisioning, etc.

Storage Virtualization Completes SAN Vision…

Where Can It Be Done?Where Can It Be Done?Host-based Host-based

VolumeVolumeManagementManagement

Network-basedNetwork-basedVirtualizationVirtualization

ApplianceAppliance Purpose BuiltPurpose BuiltAppliance (PBA)Appliance (PBA)

Controller-basedController-basedVirtualizationVirtualization

SANSAN

Intelligent SwitchIntelligent Switch

Types of Network-based SolutionsTypes of Network-based Solutions1. Intel-based platforms (Appliance):

– Simple and cost-effective form factor

– Most widely deployed approach to date

– Rich set of advanced storage applications available

– Examples: IBM SVC, DataCore, FalconStor

2. Purpose-built appliances (PBA):– Strong interoperability support in heterogeneous environments

– Support for multiple storage applications

– Designed for mid-range to mission-critical apps

– Require two management interfaces

– Examples: QLogic/Troika

Types of Network-based Solutions ContinuedTypes of Network-based Solutions Continued

3. Intelligent Switches:– Promises highest levels of scalability, availability, and

performance

– Data-center ready design

– Virtualization controller software unproven & lack key storage applications

– Relatively higher cost per port

– Switch Examples: Brocade, Cisco, McData

– Virtualization Controller Examples: EMC InVista, Incipient, StoreAge

Fundamental ConceptsFundamental Concepts• Metadata Server

– Maintains Configuration Database

– Virtualization Mapping Tables

• Control Path Software– Provides Interface Between Metadata Server and Data Path Server

– Performs Background I/O Tasks

• Data Path Software– Moves Data from Server to Storage (and vice versa)

– Performs Translation from Virtual to Physical Addresses

Note: Datapath Software is Always in the Data Path

Virtualization Architectures: In-Band vs. Virtualization Architectures: In-Band vs. Out-of-BandOut-of-Band• In-Band

– Data and Control Data are managed by same compute elements

– Initial implementations lacked scalability and performance

– Newer implementations much more scalable and “enterprise-class”

– Examples: IBM SVC, FalconStor

• Out-of-Band

– Separates data and control paths

– Requires host agents creating management complexity and potential for security breaches

– Example: StoreAge

Virtualization Architectures: Split Path Virtualization Architectures: Split Path Architecture for Intelligent Devices (SPAID)Architecture for Intelligent Devices (SPAID)

• Hybrid of In-Band & Out-of-Band Approaches

– Separates control path from the data path

– Leverages port-level processing capabilities of intelligent

switches or purpose-built controllers

– Unproven approach given the lack of availability of intelligent

switches (holds highest potential)

– Example: EMC InVista, Incipient, StoreAge

SPAID ArchitectureSPAID Architecture

Arrays

Hosts

Fabric AFabric B

Split Path Architecture Most I/O are routed through the Fast Path at near

line speeds on purpose-built ASICs Few I/Os are faulted to the Control Path (CP) No state / no cache Full fabric bandwidth Scalable across entire fabric

Intelligent FC Switch

Players and StrategyPlayers and Strategy

• IBM (SVC on Appliance)

• EMC (InVista on Intelligent Switches)

• Incipient (Incipient on Intelligent Switches)

• HDS (Internal Controller)

• QLogic/Troika (PBA via Partnerships with Software Vendors)

• StoreAge (Out of band Appliance or Intelligent Switch)

• FalconStor (Appliance)

• Cloverleaf (Appliance)

• Cisco, Brocade/McData (Intelligent Switches via Partnerships)

File System VirtualizationFile System Virtualization

• Abstraction of logical relationships above

individual file system image

• Separates user and group relationships from

specific devices

• Enables creation of exports and shares

Creates a global namespace

Server Virtualization & StorageServer Virtualization & Storage• Server virtualization presents unique challenges

for storage– Much higher SAN attach rate than traditional applications

– Shared storage & CLVM is common deployment for VM

mobility (e.g. VMotion & Live Migration)

– HBAs must be specifically qualified to run at hypervisor level

for VMWare

– Single HBA shared across all virtualized guest OSes

Server Virtualization & Virtual HBAsServer Virtualization & Virtual HBAs• Enables HBAs to be virtualized using industry-standard N-Port ID Virtualization

(NPIV)• Enables SAN best practices, standard management tools in a virtualized

environment– Examples: Fabric zoning, Array-based LUN masking

NPIV HBAs

A B C

Hypervisor

A B C

Hypervisor

HBAs

Server

OS

Single Fibre Channel

view shared among all virtual machines Logical views for each VM

or group of VMs

Server

OS OS OSOS OS

I/O I/O I/O

I/O

SAN (single view) AB

C

Example: Example: EmulexEmulex

Network & I/O VirtualizationNetwork & I/O VirtualizationIO Virtualization

– Quickly map and re-map server and storage

configuration (e.g. MAC & WWN identifiers, boot images,

etc) so that applications can be provisioned and re-

provisioned without changes to fabric or servers.

– Step toward virtual data center and stateless servers

that change personality based on demand

– Ex. HP Virtual Connect, Cisco VFrame, Voltaire

GridVision

Recommendations to UsersRecommendations to Users

• If Mostly Homogeneous, virtualization is still

important.– Stay with existing vendor’s virtualization solution.

• If Highly Heterogeneous, go with intelligent

switch, PBA, or appliance

• Ensure the Applications you want are Available

• Be Wary of Internal Controller Approaches

• Ensure Solution is Not “All or Nothing”

QuestionsQuestions

Arun Taneja

ArunT@tanejagroup.com