Storage Virtualization in the Networking Domain

1
Storage Virtualization in the Networking Domain

• David Coombs
• Chief Executive Officer
• iVivity, Inc
• Mark Ish
• Chief Technology Officer
• iVivity, Inc

2
Its High Availability!
It provides Interoperability!
Its a management tool!
Its Data Replication!
Storage Virtualization
Its a Storage Utility!
Its RAID!
3
What is Storage Virtualization

• Economic Driver
• Reduce costs without sacrificing data integrity
  or performance
• Primary technology enabler
• Networked Storage
• What it does
• Transforms storage into a utility service that is
  accessed reliably and transparently
• How its achieved
• incorporates physical disks into a large
  consolidated pool
• Virtualizes applications from the pool
• Main Benefits
• Offloads storage management overhead from
  Application Servers
• Scaling and management characteristics
• Easier to implement disaster recovery and storage
  related business continuity applications
• Easier to build open system networks and avoid
  vendor lock-in

4
Virtualization Layers
Virtualization Solutions
DB Acceleration
Server-less BU
Policy Mgmt
Security
CAS
ILM
Remote Replication
Snapshot
RAID
Interoperability
Virtualization Applications
Provisioning
Replication
High Availability
Mirroring
Virtualization (storage pooling single namespace)
Virtualization Platform
Network/Storage Protocols FCP iSCSI NDMP
NFS/CIFS Objects
Physical Storage Devices Disk, Tape Optical
5
Old Enterprise Model

• Use Host based virtualization
• Network based virtualization not generally
  accepted as ready for prime time specially for
  mission critical and larger networks
• Some penetration with IBM SVM, Data-Core, etc.
• Two options exist for network based
  virtualization
• In-band IO path gets modified in the network
  system
• Essentially a server running in the network or
  before an array
• Out of band I/O path does not get modified in
  the network system
• No intelligence in the network
• Any faults serviced by a separate meta-data server

6
In-band solution

• Implemented today in general purpose servers not
  designed for I/O efficiency
• Available Bandwidth limited by internal bus
  architectures
• Not Suitable for larger networks Latency and
  reliability concerns due to hardware and software
  layering

Virtualization Application Software
These limitations are well understood in the
networking world Why do people use CISCO and 3COM
routers instead of routing software loaded onto
general purpose servers?
Performance and reliability concerns limit use in
larger enterprise environments
7
Out of Band management
Virtualization Application Software
Virtualization Client Software

• Requires operating system specific software or
  specialized HBA to be loaded onto each server in
  the SAN
• Configuration changes typically require
  interruption of I/O to servers
• Security risk if server is added without
  specialized software or hardware
• Cannot take advantage of data-path caching
• Performance degradation in monitoring
• Limits service level agreement

Boosts Data-path performance but complicates
software development and network integration
burden
8
Pros and Cons of Old Enterprise

• Pros
• Storage networks are now a mainstay
• iSCSI is slowly but irrevocably emerging as a
  alternative interconnect due to support in
  Microsoft
• Cons
• Management is complex and expensive
• Performance scaling is not easy
• Open systems still far from a reality

9
New Enterprise Model

• Virtualization Application is specialized for the
  networking environment
• Specially designed to overcome limitations
• Application does not reside at one node but is
  distributed over the entire network
• Uses special components that facilitate key data
  modification functions in hardware at wire-speed

Integrated Control and Data management
Control path
Data-path
Data-path
Data-path

• New School
• Application distributed over the network (SPAID)
• data-path components split from control
  components
• Minimal or no host processing
• Fast as lightning
• Designed for scaling

• Old School
• One big application on Server
• Lots of host processing
• Slow as a dog
• Forget about scaling

10
Key Terms SPAID

• Split Path Array of Independent Data-streams
• Denotes separating Control from Data
• One control engine manages multiple data engines
• Benefits
• Easy to scale
• Order of magnitude better IO performance
• Order of magnitude better reliability
• Data-path can be entirely in hardware

11
Key Term FAIS

• Fabric Application Interface Standard
• a common application programming interface (API)
  framework for implementing storage applications
  in a storage networking environment.
• provides for the separation of control and data
  paths
• The control path supports the handling of SCSI
  storage and storage management command sets
• The data path provides appropriate abstractions
  for common SCSI operations
• A standard way for network applications to
  implement data-paths
• Will be an ANSI T11 Standard
• T11.5 Management workgroup
• Active since early 2004
• V1.0 draft expected this summer

12
Virtualization Methodologies
13
Developing New Enterprise Systems

• Is easier than you think
• Ability to leverage new components that
• Do lots of stuff in hardware
• Streamline development
• Eliminate up-to 60 of the effort required to
  bring a system to market
• The New Enterprise
• Cant leverage servers
• Requires developing embedded systems
• Storage Processors and Intelligent Storage
  Processors
• Linux OS a great OS for developing these systems

14
Example Component

• iVivitys iDiSX2000
• Segment leader in tech and functionality
  according to SNP report
• Leverages SPAID on chip for max performance and
  functionality
• Runs protocol processing at wire speed
• Support for virtualization data-path in hardware
• Linux embedded on chip

15
Why Linux?

• Virtual Infrastructure for Servers
• Grid Computing
• Blade Servers
• Beowulf
• MPI
• Network Connectivity Already in Place
• All Protocols Supported
• New Protocols adopted faster that any new OS
• Management infrastructure in place on servers

16
New Enterprise
Servers Linux Provides
Storage End Point Intelligence
Intelligent Switch
File and Block
17
New Enterprise

• Servers Network Intelligent
• Clusters
• Virtual Linux
• Make End Points Network Intelligent on the same
  network
• Leverage Linux for embedded Applications
• Bring Intelligence into the fabric
  (switch/router)
• Data/Control Path Separation
• FAIS (T11.5)

18
Embedded Linux Knocks

• Not fast enough in protocol processing
• Slow interrupt processing and context switch
• Will never achieve wire speed using software
• Too large for real time applications
• The ones Linux won (no more fighting)
• Open source model
• Not stable enough (not commercial product)
• Support problems

19
Embedded Linux Solutions

• Use Hardware Acceleration devices for protocol
  processing
• Use RAM disk for root file system
• Disable paging and swapping
• Provide low latency interrupt processing
• If using multiple threads use multi threading CPU
  or multiple CPUs
• Self reliance
• There is no better manual than an original
  source code

20
Interfacing to Hardware

• Use Socket interface for TOE (no lawsuit there)
• Provide a generic AF_TOE family of sockets
• Behaves identically to normal sockets but TCP/IP
  stack is bypassed
• Allow the same bypass for Storage Protocols (part
  of drivers)
• Leverage existing protocols for management
• SNIA management style (SMI-S/CIM)
• Use FAIS for Control and Data Separation

21
Software Architecture
End Customer Applications (ILM)
Servers
Integrated Backup
User
Applications (Snapshot, migration)
RAID
LVM
File Virtualization
FAIS SMI-S/CIM
Kernel
SCSI
iSCSI
TCP/IP
FCP
Linux
22
New Hardware Approach

• General Purpose Processors require protocol
  acceleration engines and a lot of glue logic
• Network Processors have no general Processing
  Power
• Look to new breed Storage Network Processors
• Protocol Acceleration built in
• Multiple General CPUs left for customer
  requirements
• Loosely coupled processor interconnect with
  specific hardware accelerated functions

23
New Enterprise Technology
The Disk Layer
The Application Layer
The Network Layer
The Storage Data Management Stack
24
Hardware Architecture
Analogous to Layer 7 Switching
High level Low Level API
FacilitatesEnd-to-End Management of

• Storage Streams
• Storage Transactions

APPLICATION ASSISTS
VIRTUAL POOLS
RAID, LBA MAPPING, AGGREGRATION
iSCSI FCP
SCSI
STORAGE I/O REISSUE
SCSI
VIRTUAL LUNS
LUN SPECIFIC I/O HANDLING
25
Software Model for iSCSI-FC