2. 2
PANASAS OVERVIEW
§ Scale-out NAS Solutions for Enterprise and Research
• ActiveStor appliances in 5th generation
• Shipping Since 2004, with 19 patents issued (more pending)
§ Panasas Management
• Storage-focused executive management team
• Highly experienced technical team
− Dr. Garth Gibson, Founder & Chief Scientist, Author of “Berkeley RAID Paper”
§ VC Funded
• Mohr Davidow, Samsung Ventures, Intel Capital, Focus Ventures,
Centennial Ventures
Industry Recognition
Faye Pairman,
Panasas CEO
Cloud Project of
the Year
Public Sector
Storage Project
of the Year
3. 3
Life
Sciences
Genomic
Sequencing
Computa1onal
Chemistry
Molecular
Modeling
BIG DATA: DESIGN & DISCOVER
Energy
Seismic
Processing
Reservoir
Simula1on
Interpreta1on
Finance
Por=olio
Op1miza1on
Credit
Analysis
Risk
Analysis
Manufacturing
EDA
Simula1on
Fluid
Dynamics
(CFD)
Acous1c
Analysis
Op1cal
Correc1on
Structural
Mechanics
Government
Weather
Forecas1ng
University
Materials
Science
Life
Sciences
Climate
Modeling
High
Energy
Physics
Private
Clouds
Defense
Intelligence
Media
Digital
Post-‐Produc1on
Rendering
Satellite
Imaging
Digital
Ac1ve
Archives
4. 4
THE SCALABILITY-RELIABILITY DILEMMA
§ Hard Drive Failures are Inevitable
• Drive failures become more frequent with scale simply because there
are more drives to fail.
• For this reason, most scalable NAS systems bring scalability that
comes at the cost of lower data reliability and availability.
§ Hardware RAID Increasingly a Problem for Legacy Storage
• Reliability and availability get worse with scale
• Multi-day rebuild times becoming common
• Odds of encountering sector errors too high, even for RAID 6
§ RAID Solutions Typically Architected for Protection, not Fast
Rebuilds
• Long RAID rebuild times lower overall data availability
• Rebuild time determined by the quantity of data to rebuild and the
RAID architecture performance
• Greater risk to production data during rebuild
6. 6
RELIABILITY AND AVAILABILITY AT SCALE
§ PanFS Delivers Unmatched Data Protection in Software
• Erasure codes protect every file individually
− No hardware RAID, no block-level rebuilds
− Limits rebuilds to affected files, not entire drives
• RAID 6+ with triple parity protection: 150x improvement over dual parity
− Guards against two simultaneous drive failures and single sector errors on
multiple drives
− Only ~25% capacity overhead and minimal performance impact vs. RAID 5+
• RAID rebuild performance scales linearly
− Mitigates risk from additional components at scale
• Randomized distribution of data means increased reliability at scale
− Odds of “one-too-many” drive failures affecting data decrease with scale
7. 7
PER-FILE RAID INCREASES RELIABILITY
§ Per-file Distribution Reduces Risk at Scale
• Small files are triple mirrored, large files are striped
• With more and more drives, three drive failures (exceeding fault tolerance)
are less and less likely to affect any given file
One file damaged;
Only need to restore File 5
No files damaged;
Can rebuild all data
File 1 (small): down 2, rebuild mirror
File 2 (small): unaffected
File 3 (small): down 1, rebuild mirror
File 4 (large): down 2, RAID 6 rebuild
File 5 (large): down 3, file damaged!
File 1 (small): unaffected
File 2 (small): down 1, rebuild mirror
File 3 (small): unaffected
File 4 (large): down 1, RAID 5 rebuild
File 5 (large): down 1, RAID 5 rebuild
x x x
Files distributed over ten drives
X = DRIVE FAILURES
x x x
Same files distributed over 20 drives
X = DRIVE FAILURES
8. 8
UNMATCHED DISASTER RECOVERY
§ Fast Time to Restore
• Restore specific files instead of entire file system
• Made possible by extra protection of namespace (directory data) in RAID 6+
§ Percentage of Files to Restore Approaches Zero with Scale
• With RAID 6+ (66% small files), a triple simultaneous disk failure means:
1 in ~200,000
files to restore
1 in ~200,000,000
files to restore
Scaling by 10x increases reliability by 1000x!
9. 9
REVOLUTIONARY AVAILABILITY MODEL
§ Extended File System Availability (EFSA) Ensures Business
Continuity
• File system remains available when other systems would be down
(even after one-too-many drive failures)
• Directory structure stays navigable and all unaffected files can be
accessed normally
• Only a small number of affected files will be unavailable until restored
from backup
§ Replaces “All Up/All Down” Model of the Past
• Targets an “always on” architecture
10. 10
PANASAS ACTIVESTOR PLATFORM
ActiveStor Appliance
§ Powered by PanFS storage operating system
§ 122.4TB per shelf à scales to 12PB
§ 1.5GB/s per shelf à scales to 150GB/s
Director
Blade
§ Metadata access
§ System management
Storage
Blade
§ Parallel data access
§ Advanced caching
Panasas Architecture
COMPUTE CLIENTS
CPUs = RAID ENGINES
PANASAS DIRECTFLOW
PARALLEL + SCALABLE
PANASAS ACTIVESTOR
§ Single global namespace
§ Single system to manage
§ Seamless scaling with no downtime
§ Pay as you grow
§ Linear performance scaling
11. 11
NEW: ACTIVESTOR 16 WITH PANFS 6.0
§ ActiveStor 16: Platform for PanFS 6.0
• 50% higher density, leveraging HGST UltraStar®
He6 6TB drives
• Faster director blade speeds RAID rebuilds
• Flash sized for RAID 6+
§ PanFS 6.0: RAID 6+ Triple Parity Protection
• Per-file, distributed RAID 6+ overcomes limitations
of traditional RAID
• Reliability increases with scale, not decreases
• Extended File System Availability ensures business
continuity when other systems would be down
• Quickly restore specific files for disaster recovery,
not whole file system
ActiveStor 16
HGST UltraStar® He6
6TB Drive
12. 12
ACTIVESTOR 16 PLATFORM
§ 50% Increase in Density
• Hitachi HGST He6 Helium-filled 6TB drive
technology
• 240GB SSD optimized for RAID 6+
• 122.4TB per 4U shelf
§ Linear Scalability
• Field proven to 12PB and 150GB/s in a single,
global namespace
− 100 shelves, 2000 HDDs, 1000 SSDs
§ Improved Director Blade Performance
• 19% faster CPU (2.53GHz quad-core)
• 4x more cache: 48GB memory
• Improves metadata, small file and RAID rebuild
performance
§ Strong at Both Throughput and IOPS
• Ideal for large file, small file, and mixed workloads
• >13,000 random 4KB file read IOPS per shelf
• 1.5 GB/sec per shelf
ActiveStor 16
Storage Blade
240GB
SSD
CPU
8GB
Cache
6TB
HDD x2
13. 13
SUMMARY
§ IT staff in technical research and enterprise environments need
scalable and reliable high-performance storage that is easily
managed
§ Selecting scale-out storage usually requires compromises
• High performance computing file systems deliver performance but
compromise on reliability, availability, and ease of management
• Enterprise NAS systems deliver high reliability and availability but
compromise performance
§ Panasas ActiveStor delivers a no-compromise solution
14. 14
PANASAS® ACTIVESTOR®
“No Compromise” Hybrid Scale-Out NAS
§ Enterprise-Grade Reliability
• Triple parity protection with RAID 6+
• Reliability increases with scale
• Extended File System Availability
§ High Parallel Performance
• Direct data access eliminates
traditional NAS bottlenecks
• Capacity and performance scale
linearly to 12PB and 150GB/s
• Flash accelerated hybrid architecture
§ Easy Management
• Minimal management effort at scale
• Set up or grow capacity in <10 minutes
• Single, global namespace
ActiveStor 16