View
2
Download
0
Category
Preview:
Citation preview
Tape in High Performance Tape in High Performance
Computing EnvironmentsComputing EnvironmentsRolf Lange SpectraLogicRolf Lange SpectraLogic
Organisations using disk-only storage
Rethinking Disk & Tape Strategies…Rethinking Disk & Tape Strategies…
Plan to start using tape
for long-term archiving 68%
Decreased
last year
13%
Plan to start using tape
for interim storage 58%
•Media &
Entertainment
•Healthcare
•High
Performance
Computing
•Life Sciences
•Surveillance
Source: Fleishman-Hillard Research for the Linear Tape Open (LTO) Program
for long-term archiving 68%
• Cost• Density• Energy Consumption• Performance• Security
• Persistence/lifespan• Portability• Reliability
“Tape has been shifting from its historical role of serving as a medium dedicated primarily to short-term backup, to a medium that addresses a much broader set of data storage goals, including:– active archive (the most promising segment of market
growth),
2011 INSIC Tape Applications System Report2011 INSIC Tape Applications System Report
growth),
– regulatory compliance (approximately 20% - 25% of all business data created must be retained to meet compliance requirements for a specified and often lengthy period), and
– disaster recovery, which continues in its traditional requirements as a significant use of tape.”
Active Archive provides an affordable, online
solution to access and store all created data.
An archive that contains production data, no
Active ArchiveActive Archive
An archive that contains production data, no
matter how old or infrequently accessed, that
can still be retrieved online. It may exist on
both disk and tape.
4
Active Archive
Extended File
System
Disk Arrays
Memory Systems
� Speed of access
� Cost
� Energy consumption
Disk Will Be SqueezedDisk Will Be Squeezed
2011 2020
Disk Arrays
Tape
� Cost
� Energy consumption
� Long term archive
� Security
• Extreme scalability for single system active archive
• Performance – Scalable drive counts for very high throughputs
– Multiple robots to support high daily cartridge exchanges
– Optional ‘enterprise’ drive support
• Reliability– Tape read reliability
– Media lifecycle management
Tape Requirements for HPC environmentsTape Requirements for HPC environments
– Media lifecycle management
– Media replication, RAIT
• System Redundancy & Serviceability– Robot, robotic interface, robotic controller, PSUs, dual port drives, global hot
spare drives, assisted self maintenance
• Storage Density
• Energy Consumption
• Cost per TB
Specifications Specifications –– Drive ModelsDrive Models
Feature IBM TS1140 Oracle T10000 C LTO-5
Capacity (Native) 4.0 TB 5.0 TB 1.5 TB
Transfer Rate (Native)250 MB/s 240 MB/s 140 MB/s
R/W Compatibility • R/W TS1130
• Reformatted TS1120
• R only T10000B
• R only T10000A
• R/W LTO-4
• R only LTO-3
Power Consumption 51 Watts 67 Watts 27 WattsPower Consumption
(No sled)
51 Watts 67 Watts 27 Watts
Interfaces • 8 Gb FC
• FICON*
• 4 Gb FC
• FICON
• 8 Gb FC
• 4 Gb FC
Library Compatibility • IBM TS3500
• IBM TS3400
• IBM TS3494
• Spectra T-Finity
• SL8500
• SL3000
• Spectra Libraries
• IBM Libraries
• Oracle Libraries
• Quantum Libraries
Media Sources Two One Multiple
MTBF 237,000 N / A 250,000
Load / Unload Cycles 300,000 >150,000 100,000
• ≈75% of HPC Market Utilizes Open Tape Technology
• ≈25% of HPC Market Utilizes Proprietary Tape Technology
Tape Drive Technology In UseTape Drive Technology In Use
LTO Tape Roadmap TodayLTO Tape Roadmap Today
http://ultrium.com/technology/roadmap.html
Announced April 14, 2010
January 22, 2010:
• “The scientists at IBM Research – Zurich, in cooperation with the FUJIFILM Corporation of Japan, recorded data onto an advanced prototype tape, at a density of 29.5 billion bits per square inch —about 39 times the areal data density of today's most popular industry-standard magnetic tape product*.
Future of Tape Capability SecuredFuture of Tape Capability Secured
industry-standard magnetic tape product*.
• “These new technologies are estimated to enable cartridge capacities that could hold up to 35 trillion bytes (terabytes) of uncompressed data*.
* http://www.zurich.ibm.com/news/10/storage.html
Density / Capacity Density / Capacity
• Tape - 4.3 PB+ per Sq. M. (Terapackdesign, TS1140 technology)• High Density NAS - 1.5 PB / Sq. M. • 4x – 5x increase in tape density in 5 years (20 TB tapes).
TerapackTerapack
DesignDesign
Areal Density of Hard Disk Areal Density of Hard Disk vsvs TapeTape
EnergyEnergy
“The disk system costs over 25 times more money to power and cool than a
similar tape system.” -Clipper Group, 2007 , (5-year cost comparison to SATA
disk)
Tape and Disk Costs – What it Really Costs to Power the Devices
“The energy cost ratio for a terabyte stored long-term on SATA disk versus “The energy cost ratio for a terabyte stored long-term on SATA disk versus LTO-4 is about 290:1.” -Clipper Group, 2008, (5-year cost comparison to D2D
backup)
Disk and Tape Square off Again – Tape Remains King of the Hill with LTO4
“The cost of energy alone for the average disk based solution exceeds the entire
TCO of the average tape based solution.” “…disk consumes 238 times as
much energy as tape under assumptions that lean toward favoring disk.” -
Clipper Group, (2010, 12-year cost comparison to average disk solution)
In Search of the Long-Term Archiving Solution – Tape Delivers Significant TCO
Over Disk
• Up to 400,000 slots in a library
complex
• Up to 8 Libraries / 16 Robots
• 40 Frames per library
Scalability Scalability -- ExascaleExascale StorageStorage
Skyway Pass-Through
connecting libraries in
complex.
T-Finity library
architecture
supports multi-
library complex.
Reliability has increased 700% over the technology
available a decade earlier
Tape Technologies Are Reliable…Tape Technologies Are Reliable…
• Advances in the coating of tape film
• Read-after-write data verification
• Error correction codes
16
• Error correction codes
• Drive technology features
simplified tape paths and servo
tracking systems
• Spectra tape libraries offer data
integrity verification
Beech, Debbie. “Best Practices for backup and long-term data retention” Sylvatica
Whitepaper. The evolving role of disk and tape in the data center. June 2009
• Tape has the best bit error rate
of any storage medium
ReliabilityReliability-- Orders of Magnitude GreaterOrders of Magnitude Greater
• “Green” tapes have debris
• Debris is typically the result of the manufacturing process – Oxide shedding– Fractured base film (mylar)– Slitting debris
What is Tape Debris?What is Tape Debris?
Enlarged area of tape section.
The large circular contamination is
about 25 μm in diameter, roughly
twice the width of an LTO4 track.
• New tapes are abrasive to the tape drive head surface and
can cause excessive wear and reduce the life span
• Debris contamination is a common cause of
performance problems:
– Poor signal performance can cause temporary data
Known Problems Caused by Tape DebrisKnown Problems Caused by Tape Debris
– Poor signal performance can cause temporary data
errors leading to read/write retries*
– Debris can migrate into data bands, obstructing reading and writing of
data causing retries or permanent data loss *
– Debris accumulation can cause tape to wind unevenly on the spool
leading to tension control problems that cause temporary errors and
retries, or may even cause tape to break*
Source: IBM LTO Media: Optimized for IBM Drives and Automation. The difference is performance
• Tape burnishing removes loose and embedded
particles and smoothes asperities in the tape
• Improves the performance level and prolongs the
life of tape head (verifies the tape surface)
CarbideCleanCarbideClean™ ™ the Spectra Solutionthe Spectra Solution
Complete Media Lifecycle ManagementComplete Media Lifecycle Management
– Very fast tape mounts of RAIT sets (up to over 2x the
speed of non-RAIT sets or non-Terapack libraries
– Protection against damaged, dropped or worn media
– Streaming rate = to number of data tapes in a RAIT set
(up to 8x a non-RAIT stream speed)
TerapackTerapack OptimisedOptimised RAITRAIT
– Available in November from HPSS and Spectra
• Dual Active-Active Robotic Transporters
• Dual Robotics Interface Modules (RIM)
• Dual Robotics Control Module(RCM) Architecture
• Global SpareTM tape drive
High Availability ArchitectureHigh Availability Architecture
• Global SpareTM tape drive
• Redundant Power Paths
• Data Integrity Verification
– PreScan
– QuickScan
– PostScan
• Tape’s purchase cost remains unbeatable for large
systems
– 10 to 15 times less expensive to purchase than disk.
• A 2.7PB tape system will have a cost of around $0.05 to $0.10 per
GB. Larger systems are even lower.
CostCost
• IT grade disk costs about $1.00 GB street price; more for
enterprise class disk
Acquisition Cost
KD1
Folie 24
KD1 Visual comparison of purchase cost of tape vs disk for a few points? 75TB, 250TB, 500TB, 1PB? Kent Dinkel; 22.10.2011
NASA Ames Supercomputing FacilityNASA Ames Supercomputing Facility
NASA Ames Case Study
Achieving Efficiency with Active ArchivAchieving Efficiency with Active Archivee
NASA Ames Case Study
• Pain Points
– Downtime due to media issues
– Maxed out data center floor space utilization
• Goals
Achieving Efficiency with Active ArchiveAchieving Efficiency with Active Archive
• Goals
– Ensure media and data integrity
– Better manage media within the library
– Improve storage density and footprint
– Have all data online, searchable and accessible
Results—NASA benefits greatly by migrating to an Active Archive:
SGI’s DMF and Spectra’s T950 (8 Frame)
– Extended file system capacity on tape
– Reclaimed 1400 sq. ft. of data center space
Achieving Efficiency with Tape and Active Achieving Efficiency with Tape and Active
ArchiveArchive
– Reclaimed 1400 sq. ft. of data center space
– Increased online archive capacity from 12 PB to 32 PB
– Increased data storage reliability
28
Results
Achieving Efficiency with Active ArchiveAchieving Efficiency with Active Archive
29
New
Storage
Footprint
• Argonne National Labs
• FMI- Switzerland
• Honeywell
• NASA Ames
• NASA Goddard
• Fermi National Labs
• Sandia National Labs
Sample HPC Customer listSample HPC Customer list
• Sandia National Labs
• Los Alamos National Labs
• Honda Research
• Colsa
• NCSA
• Howard Hughes Medical Center Research
• CERN
• Proven Innovator and History of Success
– Founded in 1979, self-funded, profitable, debt-free growth
– Intelligent integration of complete data protection solutions
– Highest customer satisfaction & support ratings in the
industry
30 Years of Success30 Years of Success
industry
• Gold Standard Manufacturer in Tape Innovation
• Manufacturing in Boulder, Colorado
High
Performance
Computing
17%Education
9%
Other
12%
• Leader in data intensive
verticals:
– HPC, M&E, Federal and Financial
– Require true enterprise solutions
Worldwide MarketsWorldwide Markets
Federal
23%
Media and
Entertainment
25%
Financial
14%• Enterprise R&D flows over to
mid-market
and department level products
• Ranked #1 in 14 out of 14 categories!
– Overall Product Ranking
– Initial Product Quality
– Product Reliability
Product Features
Awards and RecognitionAwards and Recognition
– Product Features
– Technical Support
– Sales-Force Competence
– I Would Buy This Product Again
• Ranked #1 in all 7 categoriesfor Mid-Level and Enterprise!
Thank YouThank You
34
Recommended