2. Scality RING Organic Storage 4 Jan. 2013

3. Agenda Corporate Profile Industry Trends and Business Challenges
Scality RING 4
Scality Solutions
Ecosystem
Roadmap and Vision
Conclusion

4. Scality – Quick Facts Founded 2009 Experienced management team
HQ in the San Francisco, Global reach
<50 Employees, 20 engineers in Paris
24 x 7 support team
US patents
$13M invested in Scality to-date
120% annual growth
Industry Associations
“Aggressive use of a scale-out architecture like that enabled by Scality's RING architecture will become more prevalent, as IT organizations develop best practices that boost storage asset use, reduce operational overhead, and meet high data availability expectations.”

5. A Rolling and Unstoppable Ball…

6. Industry Trends Rapid unstructured data growth
“Enterprises will see a staggering 52% growth in data over the next years — much of it an increase in unstructured data”, IDC
x8-10 in 5 years, x100 in 10 years !!
100TB today will be 1PB then 10PB
Rapid decline in storage costs
“Cost of disk storage systems is declining >30% annually”, IDC
Skyrocketing storage budget & complexity
More storage devices need to be managed
Increasing cost of commercial real estate
Increasing energy costs (power, cooling)
Cost of managing storage between 5 and 10 times the acq. cost (Gartner, IDC)
Storage Budget
TB Growth
$/TB Cost

7. Data and Storage Challenges
100s Millions of Users, 10s-100s PB of Data and Billions of files to store and serve
What all these companies have in common ?
Internet/Cloud business impossible to sustain and develop with traditional IT approaches

8. Control and Efficiency at Scale
Scality’s Mission
Their DC
Their App.
YOUR Data
Their DC
YOUR App.
YOUR Data
YOUR DC
YOUR App.
YOUR Data
Control and Efficiency at Scale
Unified Scale-Out Storage Software from primary to long term storage with Cloud’s advantages, investment protection and ready for future

9. Scality RING Organic Storage 4
File Storage
StaaS
Digital Media
Big Data*
Enterprise & Cloud System
Scale Out File System
S3 & CDMI API
Data Processing
with Hadoop
Origin Server
Scality RING Organic Storage 4
Ring Topology
P2P
End-to-End Parallelism
DATA MD
Object Storage
Replication
ARC
Erasure Coding
Geo Redundancy
Tiering
S3 CDMI
Standard
Management
MESA NewSQL DB
x86
* Available Q2/2013

10. Scale-Out Storage Software based on Shared Nothing Model
Hardware Agnostic
Racks
(10…40U)
x86 server
(1U, 2U…4U)
with
CPU, RAM, Ethernet and DAS
running Linux
Clusters
Scale-Out Storage Software based on Shared Nothing Model

11. Distributed Architecture
P2P
Scality RING
1 node
manages
1/36th of the key space
servers (6)
storage nodes
(ex: 6/server, total=36)
36 storage nodes projected on a ring
From Servers to Storage Nodes
RING Topology, P2P Architecture
Limitless Scale-Out Storage based on Shared Nothing model
Fully Distributed Storage (Data and Meta-data)

12. Distributed Architecture
P2P
DHT & Consistent Hashing
360° key space, each node manages a key range
Key generation and projection (no overhead at running time)
MIT CHORD Inter-routing Algo. + 3 Scality methods (Proxy, Balance & Rebuild)
1/6 key space to rebalance when a storage node is lost
No Single Point of Failure, No central Catalog or DB
Stateless architecture
Easy to grow, high resistance and seamless to failures
Self Healing and Organic Upgrade
Elastic Clustering
Inherent Load balancing & Symmetry
Every node can serve a request
Maximum # of hops
100 nodes 3
1 000 nodes 5
nodes 7
Topology is cashed at connector level.

13. End-to-End Parallelism
APPLICATIONS / CONNECTORS
APPLICATIONS / CONNECTORS
APPLICATIONS / CONNECTORS
APPLICATIONS / CONNECTORS
APPLICATIONS / CONNECTORS
Parallel Connectors access to Storage Nodes
Performance aggregation
Redundant Data Path
Multiple Storage Nodes per server
Minimum 6 to increase parallelism and data independence
Fast and easy rebuild
Multiple IO Daemons per Server
Control Physical and Boost IO
STORAGE NODES
STORAGE NODES
STORAGE NODES
STORAGE NODES
I/O DAEMONS
I/O DAEMONS
I/O DAEMONS
I/O DAEMONS
I/O DAEMONS
I/O DAEMONS
SSD
SATA
TIERED STORAGE
Scality Parallelism Factor
#Storage Nodes x #IO Daemons vs Simple server node with only 1 IO Engine
Independent Performance and Capacity Scalability

14. Object and Key/Values Stores
DATA MD
Object as a opaque entity with Metadata and Data
No limit for object size and number
Random block R/W access on large objects
Object Chunking, Splitting, Striping and Streaming
160 bit Key Space with 128 bit Object key
Store node acting as Simple and High Performance Key/Value stores
Fully Autonomous
Replica
Class
128 bits
160 bits
24 bits
Dispersion
Payload
Class of Service
8 bits

15. Internal Distributed DB
Scality MESA
NewSQL Philosophy
Distributed Tables across Storage Nodes
Multiple Indexes per table
Schemas updates/Flexible schemas
100% ACID
100% Elastic
Linear Transactional Performance
Automatic Fault Tolerance
SQL front-end (not exposed)
Main
DEFS
Properties/Name
Mapping Table
Key
Page ->
Page
Page #
Value

16. Data Replication No data transformation Class of Storage Rack-aware00 20 40 60 80
100 11
120
140
160
180
200
220 91
171
No data transformation
Clear/Native data format
Very fast access
Simple projection
Class of Storage
Up to 5 replicas (6 copies)
Rack-aware
Guarantee of full independent object location
Self-healing
Balance misplaced objects
Transparently proxy misplaced objects
Rebuild missing replicas
Permanent CRC of all contents (no silent data corruption)

17. ARC: Scality’s Erasure Coding
RAID 5 or 6 can’t scale without risk
With Bit Error Rate vendors information (1 bit error among 1014 bits)
55% of probability to receive an error for a volume of 10TB
99.9% for 100TB !! (worst beyond)
Replication is a great method but too costly for large storage environment
Solution: Scality Advanced Resilience Configuration (ARC)
Available as a standard RING Core feature (no extra cost)
Configuration option, running within the RING
Leveraging Reed Solomon technology

18. Better Durability & Reduced Storage overhead than Replication
Scality ARC
ARC
Fully configurable, by default ARC(14,4) meaning…
14 data fragments + 4 checksums fragments so max. loss of 4 elements in a transaction (18 total fragments committed on disk)
Read() is satisfied with minimum of 14 fragments: (14,0), (13,1), …, (10,4)
Data fragments = Native Data (no transformation)
Direct and fast read access
Scality
ARC(14,4)
Data
Check sums
Data inputs
Data source
Output to be stored
Scality RING 4 14 …
HW overhead
ratio = 1.3
(14+4/14)
(n,k) configurable to accept different tolerance profile for drive, node and site failure
Arguments against: key difference vs Cleversafe and Amplidata (only equations stored, need to decode at every read – very CPU intensive and latency added)
Better Durability & Reduced Storage overhead than Replication

19. Scality ARC Comparison
Replication A B
Storage space x3
Multiple copies
Direct access
Reduced storage space
Redundant information
Direct access
Scenario #1: loss of A
read(A+B) - read(B) = A
Scenario #2: loss of B
read(A+B) - read(A) = B
Scality ARC A B
A+B
Dispersed
A-B
Reduced storage space
Scrambled data
Latency with additional computation for each read (decoding phase)
A+B
A*B

20. Geo Redundancy
Business Continuity with “true %” availability including maintenance
Multi-site topology with Scality RING (up to 6 sites)
Replication or Geo Erasure Coding implementation – Synchronous
Or Multi-RINGs on Multi-site (independent topology) – Asynchronous
Asynchronous multiple independent RINGs
Synchronous stretched RING across 2 sites

21. 20% 80% Auto Tiering Tiering like HSM 20/80 approach Criteria
Primary Site
Site A
Tiering like HSM
Fully automated and transparent for application
20/80 approach
Perfect for Hot data on SSD/Flash and Cold data on SATA
Criteria
Age, Access Time & Object size
Potentially different data protection mechanisms
Replication with 3 copies on Tier 1
Erasure Code – Scality ARC – on Tier 2
Data Migration
across RINGs
within same site
or across sites
80%
Tier 2
Secondary Site
Site B

22. Access Methods and Standards
S3 CDMI
Open Cloud Access Strategy
Scality RING as the storage backend
File and Object Access Methods
Local and Remote
Global Namespace
Scality RS2 API (HTTP/REST)
Compatible with Amazon S3
Full CDMI server implementation
By object ID and hierarchical path- based namespace
Cloud Files as the OpenStack Swift protocol S3
CDMI
Internet
NFS
NFS
NFS
SOFS
SOFS
Scality RING

23. Management Supervisor RingSH Central management platform
Monitor application connectors & storage nodes down to individual disk drives
Passive component
Detailed Activity Log
RingSH
Command line interface
Easy to integrate or script with
Manage platform, store/retrieve/delete objects
List all keys

24. « Exceptional Performance »
“ESG Lab verified exceptional performance for an object-based storage solution, which rivals block-based solutions. Aggregate throughput scaled linearly as nodes were added to a RING. Response times improved as the RING grew in size, allowing for predictability when deploying a RING.”
Content Delivery
Object Type
Simultaneously Sustained Objects Delivered
5 servers with Intel SSD
Internet Audio (MP3)
211,424
Internet Image (JPG)
135,311
Internet Video (MPEG)
90,208
CD Audio (ISO)
18,877
Broadcast TV (HD)
2,298

25. Scality RING – The BIG picture

26. Scality Solutions Email File Storage StaaS Digital Media Big Data2
Scality RING Core P2P, Replication, Erasure Code, Tiering, API/Interface (HTTP/REST with Scality SRWS and RS2 Light, CDMI by Obj. ID), Management
Full platform integration
Certified with Openwave Messaging, Critical Path, Cyrus, Dovecot, VMware Zimbra, Open-Xchange
File Storage
Scale-Out File System for Linux
NFS/CIFS
CDMI by Path
Global Namespace
GeoSync
StaaS
Amazon S3 API
Authentication, Metering
Option Multi-Geo
Sync-n-Share1
Digital Media
Optimized SOFS for CDN Origin server
Big Data2
Hadoop integration with in-place data processing
Scality SOFS instead of HDFS
Open Cloud Access
S3 & CDMI API
1: Product available separately / 2: Available Q2/2013

27. Scality Ecosystem and Partnership
File Sync
& Share
Backup & Archive
Gateway
Hardware vendor strategic partnerships

28. Scality Product availability
Software only
Hardware agnostic
RING Core + Solutions Pack
Multi-Geo option
Price per usable storage capacity
“Appliance” based on MIS (Duo or Solo server)
RING Core + StaaS Pack
Multi-Geo option
Price per enclosure (4U)
Highest storage density on the market (72 x 4TB/4U)
Scality RING Organic Storage 4.0

29. Release Calendar & Roadmap
H1/2014 TBC
RING 5.0
“Universal
Data Platform”
- Block interface
- Policy Engine & Data Placement
June 2012
RING 4.0
- ARC (Erasure Code)
SOFS 1.0 (FUSE)
Supervisor (Keys, Agent, SNMP)
2012 |
2013 |
2014
Q4 2013
RING 4.3
OCA Phase 3 (Metering/Multi-Tenant + Swift + S3)
Act./Act. Geo RS2
Q2 2013
RING 4.2
SOFS 2.0 (NFS, CIFS) + GeoSync
Hadoop (HDFS replacement, computing on nodes)
Multipart Upload
Q1 2013
RING 4.1
Digital Media
Open Cloud Access (CDMI & SOFS)
SOFS GeoSync
Supervisor (usage)
OpenStack integration (EBS with Cinder)
Feb. 2013

30. Scality Technology Milestones
wth Replication and Tiering for primary storage (HTTP, S3)
Blob Store
Authentication, Metering, Security for S3
Policy data placement, data protection
Erasure Code data protection
CIFS based on Samba 4
Hadoop for Big Data
Cloud Files Swift Protocol
2010 |
2011 |
2012 |
Q2/2013 |
Q4/2013
CDMI
NFS
Distributed Database
File Sync-and-Share OEM Nomadesk
Scale Out File System for Linux
SQL front-end for Distributed DB
Block Storage (to add storage to VMs) with OpenStack integration via Cinder
Multi-Tenancy, QoS Mgmt (IOPS, Throughput, Capacity) & Snapshot
Mature
Ready
Beta
Alpha
Prototype
Design

31. Scality Technical Vision
Compute
Media
HPC
StaaS
Note: Underlined items are roadmap
Tenant A
Tenant B
Tenant C
Tenant D
Unified Storage Interface
(File, Block, Object and HDFS Local and Remote)
Scale Out Access Layer
Metering
Auth.
Real-time
Policy & QoS
Big Data
Analytics
Content
Indexing
Security
Storage Back-end
Local, Distant or Multi-Geo RINGs
High Perf
Transactional
Standard
Storage
1 year
Long-Term Storage
100 years
Statistics
SSD
SATA
Erasure Code SATA
Discovery
Customer IT
Infrastructure

32. Unified Scale-Out Storage Software
Conclusion
Technology Leader
Fully Autonomous
Software only solution with oem agreements
High Performance and strong Security
Proven deployed solution
Cost efficiency and rapid ROI
100% Data Availability and Durability
Unified Scale-Out Storage Software