ddn & irods · ddn & irods irods user group meeting cambridge, ma june 18, 2014 david...
TRANSCRIPT
DDN & iRODS iRODS User Group Meeting Cambridge, MA June 18, 2014
David Martin WOS Product Line Manager
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
Storage should improve collaboration
… Not make it harder
▶ Distributed, not centralized
▶ Minutes to install, not hours
▶ Milliseconds to retrieve data, not seconds
▶ Replication built in, not added on
▶ Instantaneous recovery from disk failure, not days
▶ Built in data integrity, not silent data corruption
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
DDN is the World’s Largest Privately Held Data Storage Company
Key Statistics
Overview: Providing the data backbone for the world’s most extreme, data-intensive environments – enabling organizations across multiple industries to maximize the value of their information globally. • Established: 1998 • Headquarters: Chatsworth, California USA • Over 1,000 Customers in 5 Continents
Key Industries: • High Performance Computing & Life Science • Cloud & Web Content • Rich Media • Intelligence/Federal
Go To Market: • Global Partners, VARs, Resellers
Industry Validation
World’s Largest Privately-Held Storage Co.
Fast500 Technology Company
Inc. 500|5000 High-Growth Company
Best HPC Storage Product
Best Practice for Digital Media
World Class Set of Global Customers
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
Big Data & Cloud Infrastructure DDN Announced & Pending Product Portfolio
Analytics Reference Architectures
EXAScaler™ 10Ks of Clients 1TB/s+, HSM
Linux HPC Clients NFS & CIFS
Petascale Lustre® Storage
Enterprise Scale-Out File Storage
GRIDScaler™ ~10K Clients 1TB/s+, HSM
Linux/Windows HPC Clients NFS & CIFS
SFA12KX 48 GB/s 1.7M IOPS 1,680 Drives, Embedded Option
SFA7700 12.5 GB/s 450K IOPS 60 Drives in 4U; 228 Drives in 12U
Storage Fusion Architecture™ Core Storage Platforms
SATA SSD
Flexible Drive Configuration SAS
SFX Automated Flash Caching
WOS® 3.0 32 Trillion Unique Objects
Geo-Replicated Cloud Storage 256 Million Objects/Second
Self-Healing Cloud Parallel Boolean Search
Cloud Foundation
Big Data Platform Management
DirectMon
Cloud Tiering
* Technology Demonstration
Infinite Memory Engine™
Distributed File System Buffer Cache* [Demo]
WOS7000 60 Drives in 4U
Self-Contained Servers
4
Adaptive cache with application & filesystem support for pre-staging, alignment & by-pass
S3
ddn.com ©2013 DataDirect Networks. All Rights Reserved. 5
DATA CENTER
Tape Archive
Users
Pub
lic C
loud
WOS Live Archive Storage
IME
GRIDScaler
High speed processing
Scalable storage engine
WOS Bridge
High Performance Data Center With iRODS tie-in
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
IRODS Data Grid & WOS Made for Big Data
• Hyperscale • Distributed • Collaborative
• Accessible • Secure • Flexible
DDN products assure data longevity with low latency over geographies and iRODS adds the important policy management & accessibility layers
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
WOS Key Feature Breakdown
Fully-Integrated Object Storage Appliance • 60 Drives in 4U • WOS7000 nodes support in-
node failover for ideal HA
Federated, Global Object Storage Namespace • Up to 256 nodes per cluster • Across 64 geographies, • Replicate data with smart policies for
performance and/or storage efficiency on a per-object basis
Object Disk Architecture • NOFS-no Linux file I/Os, no fragmentation, • Fully contiguous object read and write
operations for maximum disk efficiency
Latency-Aware Access Manager WOS intelligently makes decisions on the best geographies to get from based upon location access load and latency
User Defined Metadata and Metadata Search • Applications can assign their own
metadata via object storage API • Supports batch search of user metadata
DeClustered Data Protection • No hard tie between physical disks & data. • Failed drives are recovered through
dispersed data placement • Rebuilds only data, not drives
Flexible Data Protection Supports any combination of replication, erasure coding & Distributed Data Protection
REPLICATION
Self-Healing Architecture • Protection from node, Disk &
Data Center failure • Completely automated
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
Why DDN WOS & iRODS are perfect complements
WOS extends iRODS by… ▶ WOS makes iRODS more performant & scalable
• Single iCAT DB entry references multiple WOS copies • iRODS retrieves lowest latency WOS object instance
▶ WOS enables iRODS users to implement better QOS • High speed access to distributed project data • Long tail data can be stored on deep archive WOS Nodes
▶ WOS dramatically improves iRODS storage efficiency & availability • Self healing – at local node SAS bus speeds which increases data durability • Highly available – no interruption if disk, node, or site failure • WOS implements highly efficient Global Object Assure • Lowest TCO in the industry, <1 FTE for multi petabyte multi-site deployments
iRODS extends WOS by… • Integration with heterogeneous storage environments (multi-vendor disk & tape) • Additional clients and access methods • Richer storage rules and migration policies
8
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
iRODS Standard Operation
iRODS Server
JBOD ICAT DB
.. iRODS Server
SF
iRODS Server
ICAT DB
.. iRODS Server
iRODS Server
RAID
.. iRODS Server
Paris
BOS
iRODS Operation • iRules determine file store locations • iCat maintains file location metadata • All retrieves get primary file location info from
iCAT (all retrieves use primary)
Store & Retrieve Operation • iRule policy determines where file is stored
• FileA stored in BOS is replicated to SF automatically by iRODS
• iCat maintains file location metadata • File Access always reads primary copy
• FileA read in SF or Paris will go to BOS server (primary)
iRODS Zones
iRule
Replicas: BOS, SF
A
A
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
Site 1
iRODS Server
RAID ICAT DB
.. iRODS Server
BOS
iRODS Server
JBOD ICAT DB
.. iRODS Server
iRODS Server
ICAT DB
.. iRODS Server
WOS + iRODS Integration WOS Location Intelligence
SF
Paris
WOS as a iRODS Cloud resource • WOS Cluster is defined as IRODS “Cloud Resource” • Multiple WOS replicas stored once in iCAT
• iRODS utilizes WOS data location intelligence File store / retrieve operation • iRule policy specifies cloud storage method
• FileA stored in SF writes to closest node of WOS Cluster
• WOS replicates file to BOS per WOS storage policy • Single entry in iCAT DB
• File read always reads from closest WOS location • FileA read in BOS will retrieve from BOS WOS node • FileA read from Paris will retrieve from closest location
(normally Boston)
WOS Cluster
iRODS Zones
iRule
Replicas Cloud
A
A
iRODS optimizations for WOS is available today
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
Cluster “Acme WOS 1”
3
40 ms
80 ms 10 ms
Los Angeles Latency Map
San FranciscoNew YorkLondonTokyo
Power of WOS Location intelligence Minimize Latency & Enable True Collaboration
Data Locality & Collaboration 1. LA site user edits video “A”, which replicates to
Mexico City & Tokyo based on policy 2. MP Gateway immediately synchronizes iCat
metadata DB with Tokyo user 3. Tokyo user requests video “A” for processing,
WOS-Lib selects Tokyo site (lowest latency) & retrieves for the user
4. The Tokyo user extracts frames from the video & writes to WOS (new object), which replicates to Mexico City & LA & synchronizes the iCAT DB’s
Zone Los Angeles
10.8.24.101 . .
10.8.24.105
10.8.25.101 . .
10.8.25.105
10.8.26.101 . .
10.8.26.105
A A iRODS Server
WOS-Lib
Tokyo User
A
30 ms
80 ms
10 ms
Tokyo Latency Map
Best viewed in presentation mode
Los Angles Mexico City Tokyo
IRODS Server
WOS-Lib
2
Real Time Editing App
BB
A A
1
Zone Mexico City
Zone Tokyo
B
44
3
Los Angles User
Real Time Editing App
1
2
3
4
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
WOS Replication
Replication delivers both multi-site collaboration + data & disaster protection
▶ Select either Asynchronous or Synchronous replication per policy • Definable on a per-object basis
▶ Synchronous Replication replicates to remote sites to insure data protection before acknowledging a successful “write” to the application
▶ Asynchronous replication to remote WOS nodes increases performance • Two copies of data are always written to “local” (lowest latency) node before “PUT” call
returns OID to minimize/eliminate risk of data loss • Replication occurs in background, once replication successfully completes, extra local
copy is removed • Especially useful for big files and big data sets
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
WOS Local & Replicated Object Assure
Replicated Object Assure • DR for OA storage methods • Sync & Async replication supported • Optimized for large objects • Two OA replicas provide equal protection as 3
standard replicas
NDA Confidential Information
Splits files into 10 shards
WOS Replicated OA (ROA)
Application Servers
Data Center 2
Data Center 1
1 2 1 2 9 8
1 2 1 2 9 10
1 2 1 2 9 8 1 2 1 2 9 10
… 1 2 1 2 9 10
Asynchronous Replication
Writes @ LAN line speed WAN speed
ONLY on read failure
Data Center Failure
All shards stored locally
Reads @ LAN line speed
5 6 7 8
P1
P2
1 2 3 4 1
“PUT”“GET”
Client App``
WOS-Lib
WOS Local OA (LOA)
Local Object Assure: • Erasure-code based declustered data protection
• 8+2 (Data + Parity) for single copy protection • Each WOS node can withstand up to
two concurrent drive failures without loss of data or availability
• Local Data rebuilds
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
WOS Global Object Assure & Replicated Object Assure
Global Object Assure
• A minimum of 3 data centers must be deployed to provide disaster protection
• High storage efficiency • Data rebuilds from drive failures occur @ SAS Bus
speeds
Fastest LAN Path = Performance (unless DR scenario)
• Two data centers required for full DR • Store/retrieve data @ full LAN line speeds • Data replicates to remote asynchronously @ WAN
Speeds • Only retrieve data over slower WAN in DR scenarios
14
Dual encodes files & spreads across 3
data centers
Stores single data copy across 3 WOS
zones Splits files into
10Shards across 2 data centers
WOS OA Replicated
Application Servers
Data Center 2
Data Center 1
1 2 1 2 9 8
1 2 1 2 9 10
1 2 1 2 9 8 1 2 1 2 9 10
… 1 2 1 2 9 10
Asynchronous Replication
Writes @ LAN line speed
Data Center 2
Data Center 1
Data Center 3
1 2 3 1 2 3 15 16 15 16 17 18 17 18
WAN speed ONLY on failure
WOS Global Object Assure
Data Center Failure
All shards stored locally
Application Servers
The Net-Net - Ultimate customer flexibility
Reads @ LAN line speed
Best viewed in presentation Mode
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
Performance Optimized Tier
Geo-location & Collaboration (Active Archive)
Capacity Optimized (Deep Archive)
DDN-IRODS Summary Data Center View
15
WOS
IME Bridge
PFS
ES/GS
TSM / HPSS
Applications and Users
WOS
IME Bridge
PFS
ES/GS
TSM /ACSLS
Data Center 1
Bridge Bridge
Data Center 2 Additional Data Centers
Namespace Federation
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
Building a DDN & iRODS storage Grid
Seattle
Austin Tx (UT) UNC (RENCI)
Karlsruhe, DE
Replication &
Multi-site Collaboration
ddn.com ©2013 DataDirect Networks. All Rights Reserved.
Summary WOS & iRODS
WOS – iRODS Integration"• Petabyte scalability: Scale out by
simply adding storage modules"• Unrivaled Simplicity: Management
simplicity translates directly to lower cost of ownership"
• Self-Healing: Zero intervention required for failures, automatically recovers from lost drives"
• Rapid Rebuilds: Fully recover from lost drives in moments"
• Replication Ready: Ingest & distribute data globally"
• Disaster Recoverable: For uninterrupted transactions no matter what type of disaster occurs"
• File Layout: Capacity and performance optimized"
• Object Metadata: User-defined metadata makes files smarter"
UserCan Search, Access, Add and Manage Data
& Metadata
iRODSMetadataCatalogTrack data
iRODS Data SystemiRODSRule Engine
Track policies
WOS Object StoreWOS ObjectSupervisorGlobal Data Management
WOS Data Locality MgrPolicy-‐driven Replication