In this video from SC15, Larry Jones from Seagate provides an overview of the company's revamped HPC storage product line. At SC15, Seagate announced a major expansion of its HPC product portfolio including the ClusterStor HPC hard disk drive designed for Big Data applications.
Learn more: http://www.seagate.com/products/enterprise-servers-storage/enterprise-storage-systems/clustered-file-systems/
Watch the video presentation: http://wp.me/p3RLHQ-eMC
Sign up for our insideHPC Newsletter
2. 2
• Major expansion and upgrade of Seagate HPC product portfolio
• Including ClusterStor HPC HDD designed for Big Data applications
• ClusterStor L300 - Engineered Solution for Lustre
• Upgraded platform expands performance leadership for Lustre storage
• ClusterStor G200 - Engineered Solution for IBM Spectrum Scale (GPFS)
• ClusterStor platform extended to support all major HPC file systems
• ClusterStor A200 - Engineered Solution for Tiered Archives
• Tier 2 storage expressly designed for active archive and HSM applications
Seagate Announcements at Supercomputing 2015
Overview
3. 3
The Complete Portfolio for HPC, & Big Data
ClusterStor Manager
SAS SATA SSD
CP-2584
› 504 TB
› Dual Controllers
CP-2524
› 12 TB
› Dual Controllers
Spectrum Scale
CS-G200 with ISS
› 8 GB/s
› IBM SS 4.1
CS-A200CS-L300 with Lustre
› 9 GB/s
› Lustre 2.5
Object Solutions
Ceph
Scality
OpenStack Swift
› HPC Drive
› 4TB
› 10K RPM
› SMR Drive
› 8TB
› SAS SSD
› 1.3 TB
› Tiered Archive
› Up to 5 PB per rack
Cleversafe
4. 4
Vertical Markets and User Environments served by ClusterStor
Weather Healthcare Finance PharmacologyEngineeringAcademic DefenseEnergy
Data rich analytic
analysis that benefits
from increased
resolution, predictive
models and
recursive operations
to enhance accuracy
Numerically
complex,
computationally
intensive
applications
Multi-disciplinary
collaboration to
overcome extreme
problem-space,
social and economic
complexity
High fidelity
simulation, modeling
and visualization of
results
High performance
storage, networks
and computational
assets
6. 6
Next Generation Architecture Providing the Best Investment and Productivity Value
Introducing the ClusterStor L300
The Performance and Scale of Lustre
The Power of the Clusterstor
Engineered Solution
Industry’s
Fastest
Converged
Scale-Out
Platform
Highest
performance
throughput per
hard disk drive
Industry’s
highest quality
disk drives with
lowest disk
failure rate
Robust
Management
and Support
NEW!
Performance Increments
of 12 to 16GB/sec per SSU
Intel Omni-Path or
Mellanox IB EDR Network
Mixed I/O Workload Options
Improved RAS Features
Designed for
Parallel Access to
HPC and Big Data
Robust
Management
and Support
World’s Fastest
File System
Extremely
Scalable Volume
and Namespace
7. 7
ClusterStor L300 Key Components
› 5U84 Enclosure – CS2584
› SSU Haswell-class High Availability
Controller Pair
› ESU / JBOD – Up to 3 per Single
SSU (1+1 or 1+3 option)
› Mellanox Connect-X 4 Infiniband
EDR/FDR/QDR or /25/40/50/100Gb
› Intel Omni Path
›
Two (2) trays x (42) 3.5” drive slots
› Dual-ported 3.5” Nearline SAS &
SSD HDD Support
› Pair of H/A Embedded NSD
Storage Servers
› 7.2K RPM, 10K RPM HDDs, SSDs
SSU & ESU/JBOD
› Segate 2U24 Enclosure – SP2224
› Intel Haswell-class High Availability Controller Pair
› Mellanox Conect-x4 Infiniband EDR/FDR/QDR or 10/25/40/50/100Gb
› Intel Omni Path
› 24 2.5” drive slots
› Dual-ported 3.5” Nearline SAS & SSD HDD Support
L300 System and Metadata Management Unit
8. 8
ClusterStor HPC Drive: 4TB SAS HDD
0
100
200
300
400
500
600
Random writes
(4K IOPS, WCD)
Random reads
(4KQ16 IOPS)
Sequential data rate
(MB/s)
Performance Leader
World-beating performance over other
3.5in HDDs: Speeding data ingest,
extraction and access
Capacity Strong
4TB of storage for big data applications
Reliable Workhorse
2M hour MTBF and 750TB/year ratings
for reliability under the toughest
workloads your users throw at it
Power Efficient
Seagate’s PowerBalance feature
provides significant power benefits for
minimal performance tradeoffs
CS HPC
HDD
CS HPC
HDD
NL 7.2K
RPM HDD
CS HPC
HDD
NL 7.2K
RPM HDD
NL 7.2K
RPM HDD
HPC Industry First; Best Mixed Application Workload Value
9. 9
Seagate Nytro XD Cache Management Software
- Linux Filter Driver per OSS
- Monitors Writes Block Stripe Size
- Admin Definable Threshold
- Eg; 32kb Block Stipe Size or less to SSD
- Small Blocks Write to SSDs
- Data Flush/Writes to HDDs
- Large Blocks Write to HDDs
ClusterStor L300 SSU Small Block Accelerator
› Small Block Sizes are Written to the
GridRaid HDD storage pool
› The Last Accessed Small Block Stripe
is Written to the HDD OST in a
Continuous “Cache Flush” Cycle
› Small Block Sizes are Written to the
GridRaid ISS HDD storage pool
› The Last Accessed Small Block Stripe
is Written to the HDD OST in a
Continuous “Cache Flush” Cycle
ClusterStor Scalable Storage Unit
Object Storage Server #1
SSD Disk Pools are Configured as 1+1 /
RAID 10 w/OSS High Availability
Small Block Stripe
Sizes are Cached to a
SDD Disk Pool
Object Storage Server #2
Small Block Stripe
Sizes are Cached to a
SDD Disk Pool
Large Block Stripe
Sizes are written to
HDD
Large Block Stripe
Sizes are written to
HDD
10. New ClusterStor L300 Embedded Application Sever
Empty PCI Slots
With Intel Omni-Path
or Mellanox EDR HBA
New L300 Object Storage Server
- Haswell-class CPU
- PCI Slot for Network HBA
- Intel Omni-Path or Mellanox EDR
11. CS L300: New Intel Network Support “Omni-Path”
11
Top of Rack
Switches
Omni Path HBA/HFI
Inside the Laguna
Seca Controller
Omni Path Top of
Rack Switches
(eliminates single
points of failure)
12. CS L300: New Mellanox Infiniband Network Support
12
CS 9000
CS L300
Mellanox IB EDR HBA Inside
the Laguna Seca Controller
Mellanox IB EDR Top of
Rack Switches (eliminates
single points of failure)
13. 13
Block vs. File
• SFA14K 60GB/s = Raw Block
• ~45 – 50 GB/s through a PFS
Footprint
• Requires ~ 840 – 1200 drives
• = 40 – 57 RU
File Servers
• Requires ~ 8 – 12 external File Servers
ClusterStor
• Benchmarks = 1 Rack of 7 SSUs = 35RU
The Register: 11/10/15 datadirect_updates_core_storage_array; DDN slide 17: http://www.ddn.com/download/Full%20Throttle%20September%202015-DDN%20and%20Intel.pdf
Missed the ClusterStor comparison…
DDN has also announced a new product
CS 9000
63
GB/s
CS L300
112
GB/s
SFA14Ke
35
GB/s
DDN
14. Introducing ClusterStor Secure Lustre 200 (SL200)
ClusterStor
CS9000
HW Platform
Red Hat Linux Secure Edition Features
ClusterStor Client & MDS Label Checking
Multi-Level Security for Lustre Storage
• Certified Multilevel Security (MLS)
• Intelligence Community Cross Domain
• Complete and explicit audit trails
• Know who is doing what on the system
• Kerberos network authentication
• protects against insider threats
16. 16
Introducing ClusterStor G200 Spectrum Scale
The Enterprise Features and Reliability of
IBM’s Spectrum Scale File System
The Power of the Clusterstor Architecture
Proven at
Scale across
global
organizations
Easy to Use, get
up and running
in a few hours
Enterprise ready:
data protection,
management,
security and more
Industry’s Fastest
Converged
Scale-Out
Platform
Highest
performance
throughput per
hard disk drive
Industry’s
highest quality
disk drives with
lowest disk
failure rate
Robust
Management
and Support
Designed for the world’s most
data intensive workflows
Pre-integrated, tested,
tuned, ready to deploy
Accelerates workflows and
removes bottlenecks
Ensures end-to-end data
availability, reliability & integrity
Drastically lowers Total
Cost of Ownership
Automated tiering
and ILM from HDDs
to Flash to Tape
17. 17
ClusterStor Spectrum Scale – Standard
Configuration
SSU - NSD (MD) Server x 2 (SATI)
Large File Sequential Performance
- 9GB/sec per 5U84 (Clustered)
- - 5GB/sec per 5U84 (Scatter)
Meta Data Performance
26K File Creates per Second Average
2 Billion Files Capacity per 5U84
Metadata SSD Pool
~13K File Creates / sec
~ 1Billion files,
800 GB SSD x 2
User Data Pool
~4GB/sec
HDD x qty (40)
Metadata SSD Pool
~13K File Creates / sec
~ 1Billion files,
800 GB SSD x 2
User Data Pool
~4GB/sec
HDD x qty (40)
NSD (MD) Server #1 NSD (MD) Server #2
Management Server
18. 18Single Line Definition of GRIDRAID with Benefit
Fastest Rebuild for failed drive
ClusterStor GRIDRAID
Parity
Rebuild Disk
Pool #1
Parity
Rebuild Disk
Pool #2
Parity
Rebuild Disk
Pool #3
NSD
Server
Parity
Rebuild Disk
Pool #4
Traditional RAID
Parity
Rebuild
Disk
Pool #1
NSD
Server
GridRAID
De-clustered RAID 6: Up to 400% faster to repair
Rebuild of 6TB drive – MD RAID ~ 33.3 hours, GridRAID ~ 9.5 hours Recover from a disk failure and return to full data protection faster
Repeal Amdahl’s Law: speed of a parallel system is gated by the performance of the
slowest component Minimizes application impact to widely striped file performance
Minimize file system fragmentation Improved allocation and layout maximizes sequential data placement
4 to1 Reduction in NSDs Simplifies scalability challenges
ClusterStor Integrated Management CLI and GUI configuration, monitoring and management reduces Opex
Feature Benefit
19. 19Best of Most Widely Adopted Parallel File System Solutions
Choice in High Performance Storage
Performance Efficiency
Highest performance throughput
per hard disk drive
Engineered Solution
Pre-integrated, tested,
tuned, and shipped
ready to deploy – built-in
rock solid productivity
Reliability
Industry’s highest
quality disk drives
with lowest disk
failure rate, mitigates
root cause to
degraded
performance
Scalability
Sustained linear performance,
maximize production-level
uptime, keep HPC initiatives on
schedule and on budget
Robust Management and
Support
Comprehensive file system
management , RAS/Phone home,
holistic hardware monitoring with
health alerts
Five
Unique
Solution
Values
1
Lustre®IBM®
Spectrum Scale®
2
34
5
21. Seagate Confidential 21
ClusterStor A200 Active Archive Product Overview
Active archive object storage tier for ClusterStor
Lustre platform
Combined with Clusterstor HSM to provide automatic
policy-driven data migration & retrieval
Optimised for a balance of cost, performance &
density
Utilizes network erasure coding to provide high levels
of data availability and data durability
No single points of failure, no single points of
maintenance
10Gb Ethernet node connectivity
Lustre 2.5.x
ClusterStor 9000
HSM
Packaged as upgrade to ClusterStor
CS A200
ClusterStor A200
22. Seagate Confidential 22
ClusterStor A200 $/GB
Relative usable storage costs vs Object & Tape
ClusterStor A200 Archive Tier Value Proposition
› Cost effective, deep & fast
o Single Controller per SSU
o 8TB SMR SATA HDDs
o Increase research productivity
› Improves data availability via network erasure coding
Value Proposition vs tape- CS A200 as target for active data
› Much better accessibility (<1s vs many minutes)
› Better data availability
› Better data durability
› Better performance
For active archiving, CS A200 accessibility & performance
benefits more than justify the additional cost vs tape
ClusterStor CS A200 Tape
Relative Price per GB
23. Seagate Confidential 23
CS-A200 Economics, Performance, & Availability
Getting Object Storage Right
Economics
› Single controller per SSU
› Utilize most cost effective
drives - 8TB+ SMR SATA
hdds
› Cost efficient erasure coding
Performance
› Parallel read/write access paths
› Reduced filesystem
fragmentation increases
research productivity
› Read/write throughput of 10GB/
s per rack (7ssus)
Availability/
Durability
› 11 “9’s” availability (12+SSUs)
› No single points of failure, no
single points of maintenance
› Improve data availability &
durability via network erasure
coding
24/7
24. Seagate Confidential 24
ClusterStor A200 Data Layouts
10Gbe Networks
Data Mover
Archive Software Library
Customer Application
Archive Software Library
8 Data Chunks 2 Parity Chunks
› 8+2 Declustered Erasure Coding
• Objects are sharded up into 8+2 data+parity chunks
& written across 10 networked SSU’s
• Data resiliency with up to two concurrent SSU
failures
› Highly parallelized data access
• 2x 10GigE ports per SSU
• Up to 10 GB/s per rack throughput
› Highly available and durable configuration
• Fast data rebuilds (8TB data rebuild in less than 1 hr)
› Seamlessly scalable
• Performance, capacity, & object counts
› Upcoming Capabilities
› x+y erasure coding options
› Multiple, flexible data layouts
CS A200 Network Erasure Coding
25. Seagate Confidential 25
Key HSM v1
features
Policy-driven migration from primary storage platform / tier to lower
tier/target storage
› Age, last read, size etc
Policy-driven stubbing/release of payload to free up storage
Automatic retrieval/rehydration of file when stub is accessed
› Rehydration of complete object
HSM v1 supports ClusterStor Lustre filesystem/ platform
› Spectrum Storage Platform (GPFS) planned for next release
ClusterStor HSM v1 Functionality Overview
ClusterStor A200
26. Seagate Confidential 26
Clusterstor A200 Tiered Archive Features
ClusterStor A200
Storage optimized for cost per GB
Global namespace
“infinite” numbers of files
HSM API - support for
Lustre, GPFS, others TBA
Performance much faster
than tape (up to 10GB/s per
rack)
High density storage (up to
3.6PB usable per rack)
Can achieve 5 “nines”
system availability
Rapid drive rebuild (<1hr for
8TB in a large system)
Integrated management and
consensus based HA
X+y network
erasure coding