sgi solutions - eresearch australasia 2016 conference · 1/11/2012 · 1 ©2012 sgi sgi solutions ....
TRANSCRIPT
1 ©2012 SGI
SGI Solutions In the Era of Data-Intensive Science Jill Matzke, PhD Director, High End Servers
2 ©2012 SGI
Big Data Buzz
•Is it really new?
•Is it really that big?
•Is it really that hard?
2
HPC: Mapping, Reducing ror Years
16GB
3 ©2012 SGI
• Personalized Medicine
• National Security
• Social Sciences • Business
New Users, New Use-cases New Computer Scientists
3
The Stakes can be VERY High
4 ©2012 SGI
=> New Imperatives
• Lower HPC Complexity
• Fast Algorithm Prototyping
• Real-Time Results
5 ©2012 SGI
Meeting these Imperatives Across the Data intensive workflow
Ingest Crunch Analyze
Fast Data Access
Safe, Efficient Archive
6 ©2012 SGI
Ingest Crunch Analyze
Fast Data Access
Keep Data Safe, Economically
SGI Hadoop Clusters
Meeting the Imperatives Across the Data intensive workflow
7
SGI Hadoop Clusters: Lower Complexity => Fast Time to Results
8 ©2012 SGI
• Flexible, optimized and specific to customer requirements.
• Performance
• Power
• Density
• Cooling
• Storage options
• Price
SGI Hadoop Clusters Designed, Integrated to order
9 ©2012 SGI
1/2 Rack: 128 TB useable capacity Multi-Rack:
Petabytes useable capacity
10GigE 1 Rack: 256 TB useable capacity
Import, Export, Search, Mine, Predict & Visualize data for Business Intelligence
• Purpose designed and built
• Performance optimized
• Factory integrated
• Cloudera certified
• Power managed
SGI Hadoop Starter Kits
10 ©2012 SGI
SGI Hadoop: Proven • Leading commercial and US
government supplier
• Deployments 40,000+ nodes Individual clusters 4,000+ nodes
11 ©2012 SGI
Meeting these Imperatives Across the Data intensive workflow
Ingest Crunch Analyze
Fast Data Access
Safe, Efficient Archive
12 ©2012 SGI
Ingest Crunch Analyze
Fast File Access
Fast, Eocnomical Archive
SGI UV
Meeting the Imperatives Across the Data intensive workflow
13
One Platform: Many Advantages
• Lower Complexity
• Rapid Prototype
• Real-time Results
14 ©2012 SGI
SGI UV
• Focus onYour Science, Not IT Problems – Single-system to 4096 Intel E5 cores
• No-Limit Computing, Built on Industry Standards – Runs off-the-shelf Linux
• World's Largest In-Memory System for Data-Intensive Applications – 64 Terabyte cache-coherent memory
World-leading Capability for Data Intensive Work
14
100s Systems Shipped, 1000s Users
15 ©2012 SGI
Modular Design, Configuration Flexibility Supports GPU, Intel MIC
SGI UV Start small and grow … or start big.
16-128 core 32GB-4TB
64-512 core 256GB-16TB
256- 4096 core Up to 64TB
UV 2000
UV 20 16-32 core 32GB-1.5TB
15
16 ©2012 SGI
SGI UV 100s Times Faster than Flash
Standard Rackmount Server 1.2TB High End flash
Bandwidth (R/W): 2.5-3.0GB/s Latency: 15-47 microseconds
Source: FusionIO.com
100X Performance 35X Price/Perf.
UV 2000 1TB memory
Source: SGI Benchmarks
Bandwidth (R/W): 236 GB/s Latency: 0.1-0.5 microsecond
16
17 ©2012 SGI
SGI UV Leave the node memory limits of scale-out computing behind.
17
“..significantly enhance the capabilities of the NSF to see and understand large volumes of data…” Oak Ridge Nat’l Labs
“SGI UV frees us from memory constraints.” Human Genome Center, U Tokyo
18 ©2012 SGI
SGI UV Rapid innovation: Invent on your laptop, scale on SGI UV, no re-write required.
SGI UV
Scale-out Systems Develop Decompose Messaging Scale Reassemble
Develop (PC) Scale Next Idea …
18
“…unparalleled ease of use for rapidly testing new ideas … dramatically increasing users’ productivity.” Pittsburgh Supercomputing Center
Next Idea …
19 ©2012 SGI
Global Sentiment via Wikipedia
19
• 42 Million
Dates in the Past Millenium
• 80 Million Locations
• 24 Hours Development Time
sgi.com/go/wikipedia
20 ©2012 SGI
SGI Solutions in the Era of Data Intensive Science
Ingest Crunch Analyze
Fast File Access
Fast, Eocnomical Archive
SGI Infinite Storage DMF
SGI MAID - Arcfiniti
21
Transactional, Persistent Data
• Lower Complexity
• Fast Scalable Access
• Efficient ‘Zero Watt’ Disk
22 ©2012 SGI 22
Real-world data => Data Silos
23 ©2012 SGI 23
In the ideal: All Data Always Available in Time
24 ©2012 SGI
Challenge: Different Data Needs Different Storage
SGI Shipped over 500 PB this Past FiscalYear
25 ©2012 SGI 25
DMF: Automating storage tier virtualization Content & Metadata Modify, Collaborate, Archive Route & Reuse
26 ©2012 SGI
DMF: Automated, Policy-Based Tier Virtualization
26
DMF: Automating storage tier virtualization
27 ©2012 SGI
SGI MAID – Archive with ‘in-time’ Access Zero-Watt Disk
Disk-Based Core Platform – To 2.6PB raw storage per cabinet
Only System with Deterministic savings in power and cooling
– All disks are powered off when not in use . – 50-75% power savings – Maintains Whole-Array Access
Multiple System “Personalities” – Native MAID: ideal for HSM, D2D and archive – VTL: reliable, high performance target for backup
27
28 ©2012 SGI
ArcFinitiTM: Seamless Access to Data
• Feed many apps simultaneously
• Compatible via NFS or CIFS • Integrated HSM: SGI DMF • Disk/file-based archive for
fast, secure access to any data
MAID + DMF
29 ©2012 SGI
SGI Meeting the Imperatives For Data Intensive Science
Thank You!