Mike Mesnier

1
Object-based Storage 101

• Mike Mesnier
• January 2003

SNIA
2
Storage Trends/Facts

• Storage is moving to the network
• Servers connect to consolidated NAS or SAN
  storage
• Storage and networking fabrics are converging
• iSCSI, iFCP, RDMA/TCP,
• Storage is becoming more autonomous
• Standards for richer interfaces and protocols
• More functionality in the storage device
• Research focusing on intelligent storage
• Self-aware, self-managing, self-configuring

3
Evolving Data Center
Compute and storage platforms are tailored for
specific functions.
Switched Fabric
Building Blocks
Storage Bricks
Compute Bricks
SAN Storage
4
Storage Utopia ?

• Imagine a storage brick that is
• Always available and reliable
• Self-securing
• Self-managed
• Infinitely scalable (i.e., stackable)

Object-based storage is a step closer
5
Agenda

• ? Storage today
• Emerging solutions
• Object-based storage
• SNIA activity

6
Architectures Today
Local Storage
Shared Data
Shared Capacity
How did these evolve?
7
The Trade-offs

• DAS physically secure and simple
• Not scalable
• Limited connectivity for capacity sharing
• No data sharing
• SAN scalable capacity sharing
• Limited (coarse) security through switch
• No data sharing
• NAS secure sharing (data or capacity)
• Limited scalability

8
The Usage Models

• DAS and SAN provide raw block storage
• No model for the stored data (e.g., files)
• No support for data sharing (e.g., concurrency
  control)
• Storage applications built using DAS or SAN
• File systems, databases, video servers, etc.
• App. implements a data model (e.g., a file)
• May implement direct data sharing (e.g., via
  clustering)
• NAS is an application for indirectly sharing data
• Servers export local file systems over the
  network
• Clients share data by sharing files

9
Challenge Data Sharing

• Data centers need to share data.
• Backup, HSM
• Clustering for scalability (e.g., file server and
  db)
• Sharing data directly on the SAN is difficult.
• Concurrency control, lock management, versioning
• Distributed security
• Sharing data through NAS is easy
• But imposes limits in scalability
• Single point of failure bottleneck

10
Agenda

• Storage today
• ?Emerging solutions
• Object-based storage
• SNIA activity

11
Emerging Solutions

• Virtualization (NAS SAN)
• Aggregates heterogeneous NAS and SAN devices
• Clustering the file system
• Mitigates file server bottleneck
• Giving clients direct access to storage devices
• File servers share block metadata with clients
• Eliminates file server bottleneck
• Changing the device interface
• From blocks to objects

12
1st Generation File Server
OSD TWG
BLOCKS
NAS Islands
SAN Islands
13
2nd Generation File Server
Clients
Must be trusted
METADATA
Servers
DATA
Trusted SAN
Eth switch
MANAGEMENT
Difficult to directly share
14
Agenda

• Storage today
• Problems and partial solutions
• ? Object-based storage
• SNIA activity

15
Object Storage

• An object comprises
• Application data (e.g., file, record)
• Device-managed metadata (e.g., block allocation)
• User-accessible attributes (e.g., access times)
• Objects have file-like methods for access
• Open, close, read, write, get/set attributes
• Commands are authorized
• Object-based storage devices
• Disk drive, appliance, controllers

ID x123 Blocks3,42 Length512
Objects can be self-describing!
16
Object Storage Model
Object Interface
Block Interface
17
So Whats the Real Value of Objects?
18
The Value of Objects

• Better security via capabilities
• Each object can have its own security domain
• All I/O is authorized by the device
• Easier to share data
• Files and records can be stored as objects
• Low-level metadata managed by device
• Opportunities for intelligence
• Attribute-based learning for resource allocation
• Better caching, pre-fetching and staging of data
• Self-configuring storage w/ continuous
  reorganization
• Layout objects to best serve client requests

19
Value 1 Security

• Separates policy from enforcement
• Storage managers set policy
• Storage devices enforce the policy
• Prevents unauthorized access
• Minimizes interaction with storage manager
• Minimizes state kept on device
• For better scalability, recovery and cost

20
Security Types

• Types of security
• Authentication you are who you say
• Authorization you have permission
• Integrity data is not corrupted/modified
• Privacy data is not seen
• TWG is considering two scenarios
• Channel is trusted
• Channel is not trusted

OSD handles all but authentication.
21
Preventable Attacks

• Snooping or modification of commands and data
• Unauthorized access via modified capability
• Delay and replay attacks
• Guards against these attacks, respectively
• Transport or app-level encryption for privacy
• Transport or app-level digests for integrity
• Cryptographically secure capability
• Nonce (timestamp) attached to each command
• Requires loosely synchronized clocks
• Only needed when channel is not trusted

22
Value 2 Data Sharing

• Less metadata to keep coherent
• No block allocation or free block lists
• Shorter lists to manage
• Objects may contain aggregation metadata
• Leads to better scalability
• Backup and HSM
• No need to co-locate w/ application (e.g., file
  system)
• Only backup necessary objects (not entire volume)

OSD is really a file system less the naming
23
3rd Generation File Server
Clients
SAN
Eth switch
Managers
MANAGEMENT
Intelligent Device Space ManagementBackup/Recover
yQoS via attributesSecurity
Object-based Storage Devices
24
Value 3 Intelligence

• Objects can have rich attributes
• Timestamps, accounting information, QoS
  parameters, group and user information,
  client-specific usage patterns
• Many opportunities for policy-based management
• E.g, attributes may describe backup and QoS
  policies
• Attributes may also provide information gain
• An objects actions may be correlated with its
  attributes
• E.g., Any object written within the last 24 hours
  will be read up at 4 a.m. the following day
  (i.e., for backup)
• Can devices actually learn??
• This is an active research topic
• OSD facilitates a further investigation
• Big questions which attributes really matter?

25
Agenda

• Storage today
• Emerging solutions
• Object-based storage
• ? SNIA activity

26
The SNIA TWG for OSD

• Charter and Program of Work
• Requirements for OSD
• Transport independent definition of OSD
• SCSI Standard for OSD
• White papers industry demonstrations
• 100 members and over 30 organizations
• academia, industry, National Laboratories and
  startups.

27
Work Items in the TWG

• Establishing the commands v1 DONE
• Establishing the attributes v1 DONE
• Security architecture v1 DONE
• How should we identify and locate objects?
• How to maintain integrity through failures?
• Should we support transactional semantics?
• Management of hundreds or thousands of OSDs.

28
OSD Functions

• Basic Protocol
• Read
• Write
• Create
• Delete
• Open
• Close
• Get
• Set
• Append
• Clear

• Security
• Authorization
• Integrity
• Privacy
• Attributes
• Aid to file systems
• Hints to device
• QoS and Priorities
• Logging Statistics

29
Summary

• A big challenge in the enterprise is data sharing
• Must be secure and scalable
• Objects complement existing technologies
• Add security, low-overhead data sharing and
  intelligence
• Enable the 3rd generation file server
• OSD is more active now than ever
• SNIA has become the focal point
• Will complete v1 SCSI standard 1st half of this
  year.

30
Call to Action

• How must your file system change to support OSD?
• How will OSD change your backup story?
• See the latest OSD standard at T10
• Understand our plans for other transports
• Get involved in the OSD TWG
• Architecture
• Standards efforts
• Industry demonstrations

31
Further References

• Academic research
• www.pdl.cmu.edu
• www.dtc.umn.edu
• csl.cse.ucsc.edu/obsd.shtml
• Standards work
• www.snia.org/osd
• www.nsic.org/nasd
• www.t10.org/scsi-3.htm (see OSD)
• Industry research development
• www.intel.com/labs/storage/osd
• Download OSD Reference Code
• www.haifa.il.ibm.com/storage.html
• www.lustre.org
• www.panasas.com

32
Our Leaders

• Mike Mesnier, Julian Satran (co-Chairs)
• Applications Erik Riedel
• Education Tom Ruwart
• Management Ken Samarra
• Security Michael Factor

Contact us to get involved!