Transcript
Object Stores in Distributed Research Data Management
1
DDN SFA Storage Appliances for demanding data environments
SFA12K‐40 Performance Summary The World’s Fastest HPC Storage Foundation
EXAScaler
Write
Read
Raw Device
32.6 GB/s
39.9 GB/s
EXAScaler (obdfilter)
28.5 GB/s
33.4 GB/s
RAID 6; DirectProtect DIF: ON; ReACT: On; 1MB IOs
GRIDScaler
Write
Read
Raw Device
32.6 GB/s
39.9 GB/s
GridScaler (IOR)
32.3 GB/s
35.6 GB/s
RAID 6; DirectProtect DIF: ON; ReACT: On; 4MB IOs
In-Store Processing Virtualization DDN Hypervisor Minimizes Latency & Saves on TCO InfiniBand Client Ports Ethernet Client Ports
Dedicated I/O Bridge
Back-End SAS HBAs
Application Memory
Dedicated I/O Bridge
Multi-core CPU Application Processor (AP)
Multi-core CPU RAID Processor (RP)
Many Ways To Save: • Data Center Space • Latency
High Speed Bus
File Server
File Server
File Server
Dedicated PCI-e I/O
……
Virtual Disk Block Driver
• • •
Cache Memory
Memory Pointers (Virtual Disks)
Multi-Threaded Rea-Time RAID Engine, Hypervisor
File System Licenses Management Overhead Networking
SS8460 – Highest Density Enclosure
84 Drives – SSD, SAS, SATA ‐ in 4 rack units
Up to 336 TB
SFA7700 The First Hybrid Flash Storage Appliance with Application-Awareness Hybrid Hybrid Controller with SFX Flash Cache
Application‐Aware Data Management
Solid State performance with HDD Economics
SFX API Accelerates Data & Metadata
Efficient
10GB/s • Up To 600K IOPS Available 1H12
Smart
Simple
SFX Minimizes Disk Investment
Fully‐Integrated Modular Appliance
Industry‐Leading Data Center Efficiency 396 Disks in Only 20U
DirectMon Enterprise Appliance Management 6
DDN WOS The World’s First Enterprise Object Store for Web Scale Computing V2.5 Shipping End Q412
WOS Core is a simple storage system for objects • Today, the complexity of getting bits onto disks is preposterous • Starting from zero, DDN redesigned this and launched 3 years ago • No conventional extents‐based filesystem • No conventional disk RAID management • Single disk seek
WOS Core API The Foundation for All WOS Connectivity WOS Container
WOS Cloud Latency Map & I/O Routing
Object ID: ACuoBKmWW3Uw1W2TmVYthA
80 ms 10 ms 40 ms
WOS Security Signature
WOS Client
Replication/Protection Policy Async, Sync, Erasure Coding
64‐Bit Checksum User Metadata (<64MB)
WOS Library (WOSLib)
• • •
Key Value or Binary Tag = Beach; Thumbnails
Full File/Object >1MB Objects Segmented into 1MB Chunks
• •
Client‐side library for Linux Apps Applies extended object attributes Maintains object access latency & location awareness across the cluster, routes requests to the least latent path Delivers highest performance access Simplified instruction set •
PUT, GET, DELETE object, RESERVE ObjectID, etc
9
WOS ObjectAssure Single Copy Local Data Protection WOS Object Assure • • •
Client App ``
WOS-Lib “PUT” “GET”
• •
1
2
3
4
P1
5
6
7
8
P2
1
Works with both WOS‐Lib & REST API’s OA operates within a single WOS node OA is enabled by specifying a single (1) replica in a WOS storage policy OA & replica storage methods can be mixed inside a WOS cluster OA detects concurrent multi‐disk errors & corrects for 2 separate concurrent disk errors on a per‐WOS node basis
Object Latency (ms) – THE BEST Object Storage Latency With Replication or ObjectAssure – WOS Wins objsize 4KB 50KB 500KB
put/get get get get
2-repl 9.1 12.7 38
ObjectAssure 8.2 16.2 35.1
4KB 50KB 500KB
put put put
24.7 37.1 64.3
39.5 47.9 56.9
Best viewed in presentation mode
10
DDN |
® WOS
Enabling Real‐Time Global Collaboration Web‐Scale, High‐Performance Cloud Storage Appliances 99% Efficiency, Petabyte‐Class Peer:Peer Technology
Connectors
Android & iOS Client WOS API WOS NAS Access OID management HTTP, C++, Java
Desktop Client
WOS Cloud™ Multi‐Tenancy WOS Core [Peer:Peer Object Storage]
WOS Cluster Management 11
API: S3 Client
Limitless Scalability Eliminates the limitations of traditional file systems
Store Objects Intelligently User‐Defined Metadata allows customers to understand their data
Latency‐Aware Access Manager
Global, Peer:Peer Self‐Healing Object Storage Clustering WOS Policy Engine Magic box of objects Replication Engine
ObjectAssure™ Erasure Coding
De‐clustered Data Management
6/8/1 2
Distribute data across 100s of sites in one namespace
Self‐Healing Intelligent Data Management system recovers from failures rapidly and autonomously
Next Gen. Research Data Management App/Device ``
Parallel Filesystem Clustered NFS/CIFS
•
HPC Cluster
• •
Flexible, Extensible metadata management ‐ iRODS delivers a powerful database for managing searchable metadata Distribution/Collaboration Ready ‐ Based on simple policies Extremely high scalability‐ add further systems to the network in any location for additional scalability 12
Conclusion • Administrators! – Stop managing ext filesystems! – Stop managing RAID volumes! – Stop using fsck/lvm/growing/shrinking filesystem! – Stop working out how to synchronize multiple distributed filesystems! – Do Spend your time on higher level functions: • Getting better value out of your data using better metadata management and search techniques • Developing advanced policy‐based management
