Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
IBM Systems2
© 2019 IBM Corporation
High level concept of Media Platform
Ingest Editing Color Grading Finishing
Post Production
3
Off-site Work
クラウドオブジェクトストレージ
x
CloudBusinessPlatform
Remote Sites
Multi-Cloud Environment
Aspera
CloudObject
Storage
Aspera
Aspera
Watson Media
Cloud
ICOS
Tape
Archive
3rd party systems
FlashFlashFlash
High Performance Parallel Filesystem
Tape
HSM
System Storage
Orchestration & Workflow Management
Orchestration: AREMA/Aspera
MediaCloud
Service
MediaCloud
Service
Reference:Concept from discussions with Japan customers
4
OrchestratorCarrier & Engine for Jungles of App
DatahubDams & maps for Ocean of Data
Disk TapeFlash
Compute
& Storage
Software-
Defined
Infrastructure
Applications& Tools
CPU GPU
Framework
& Libraries
Clinical RWE
High Performance Data & AI (HPDA) Architecture
ImagingGenomics
5
Carrier & Engine for Jungles of App
Dams & maps for Ocean of Data
Disk TapeFlash
Compute
& Storage
Software-
Defined
Infrastructure
Applications& Tools
CPU GPU
Framework
& Libraries
Clinical RWE
High Performance Data & AI (HPDA) Architecture
ImagingGenomics
IBM Storage and SDI
© Copyright IBM Corporation 2018
AI Data Pipeline with Spectrum StorageImproved data governance with storage offerings for end-to-end data pipeline
Spectrum Scale
Cloud Object
Storage
Cloud Object
Storage
Elastic
Storage ServerElastic
Storage ServerElastic
Storage Server
Transient Storage
Global Ingest
Fast Ingest /
Real-time Analytics Archive
Spectrum
Archive
Hadoop / Spark
Data Lakes
Data In
Insights Out
INSIGHTSANALYZE / TRAININGESTEDGE CLASSIFY / TRANSFORM
SSD
SDS/Cloud
Cloud
SSD/Hybrid
Hybrid/HDD
TapeHDD Cloud
Trained Model
SSD/NVMe
ML / DLPrep Training Inference
Spectrum DiscoverElastic
Storage Server
Cloud Object
Storage
Elastic
Storage Server
ETL
Classification &
Metadata Tagging
Inference
IBM Systems
Where it all began
• Tokyo Metropolitan Government VOD Trial
• Interactive TV for new "borough" of Tokyo
• Applications: movies, news, karaoke, education ...
• Video distribution via hybrid fiber/coax
• Trial went "live" June '96• 500 subscribers
▪6 Mbit/sec MPEG video streams
▪100 simultaneous viewers (75 MB/sec)
▪200 hours of video on line (700 GB)
▪12-node SP-2 (7 distribution, 5 storage)
GPFS use cases more than 20 years !
IBM Systems8
IBM Systems
Kunde sucht hochperformaten lokalen geteilten Speicher
9
Different operating plattforms and
applications share one file in the
Spectrum Scale file system
Parallel reads and update from
different operating environments
LAN
App App
Single extraordinarily largeName Space
ESS1 ESS2
S i n g l e N a m e S p a c e
Hadoop
Posix Filesystem Spectrum Scale ( GPFS )
Object
file
Spectrum Scale
Spectrum Scale
Spectrum Scale
Spectrum Scale
SpycerNode – Media Storage
Contact: [email protected]
Declustered RAID /
Erasure code →
Highest availabilty
at all times
No performance
degradation over
time
Instant file response
with up to 8 Gbyte/s
and million IOPS
system throughput
Suitable for any
requirement due to
multiple configuration
options
Unified file and block
storage → High
deployment flexibility
Challenges of the
Broadcast & Media industry
→ Media workflows require low latency
file response
→ High system throughput due to large
media files
→ Extremely valuable content requires
high availability and reliability at all
times
How SpycerNode in combination with
Spectrum Scale addresses these challenges
IBM Systems12
PoC with IBM Business Partner SVA and Tier-1 supplier shows very encouraging results
Proof of Concept for Tier-1 supplier
We demonstrated
our ability to
decrease HiL
testing time more
than a third vs
EMC Isilon NAS!
Anbindung von Simulationseinheiten ( HiL)
IBM Systems
• Spectrum Scale allows transparent migration of data
among Flash and HDD storage tiers
Kunde sucht
günstiges Archiv für große Datenmengen
• Spectrum Archive can be configured as tape tier
• IBM Cloud Object Storage (COS) can be configured as an object / cloud tier
• Spectrum Scale policies can be used to automatically migrate files from disk to tape or to IBM COS
• After migration, the file or object remains visible in Spectrum Scale and can be recovered
Business Challenge
To enhance healthcare,
Sira supports initiatives
such as the Qatar
Genome Programme,
which aims to deliver
personalized care. How
could Sidra facilitate
research into thousands
of genome sequences?
Gain
ready access to
bioinformatics data
Reduce
time-to-completion for
hundreds of thousands of
analytical jobs by
maximizing resource use
“Our goal is to build a unique and integrated
platform to help researchers analyze their
bioinformatics data in an easy and efficient
way.“
Shafeeq Poolat, IT administrator
SidraSupporting game-changing genomics research to improve the health of a nation
Geography: MEA
Industry: Education
IBM Solution Components:
IBM Spectrum Scale™, IBM Spectrum Archive™,
IBM Storwize ® V7000, IBM FlashSystem ®
Case Study
Video
IBM Systems
Spectrum Archive (aka LTFS) als letzte Speicher Tierstufe im Filesystem
15
LTFS Library Edition
Linux Server
Tape Library
NFS / CIFS
ArchiveManagementSolutions
Application file access to tape
Spectrum ScaleElasticStorageServer
File system
LTFS Phase 1LTFS Format EnablementSingle Drive Support (2010)
LTFS Phase 2 Digital Archive EnablementTape Automation Support (2011)
LTFS Phase 3 - SDSIntegrated Tiered Storage Solutions (2013)
Application file access to tiered storage
Spectrum Archive
Anlegen und Schreiben von Dateien – Placement Policy
16
/home/appl/data/web/important_big_spreadsheet.xls
/home/appl/data/web/big_architecture_drawing.ppt
/home/appl/data/web/unstructured_big_video.mpg
/home
/appl
/data
/web
/home/appl/data/web/important_big_spreadsheet.xls
/home/appl/data/web/big_architecture_drawing.ppt
/home/appl/data/web/unstructured_big_video.mpg
Policy EngineGlobal Namespace
Note: Alle 3 Dateien imgleichen Directory, aberjede in einem anderen
Storage Pool
Tier 1: SSD/Flash Tier 2: SAS drives 15k Tier 3: 14TB SAS-NL drives
Lo
gic
al P
hysic
al
Migration Policies
17
/home/appl/data/web/important_big_spreadsheet.xls
/home/appl/data/web/big_architecture_drawing.ppt
/home/appl/data/web/unstructured_big_video.mpg
/home
/appl
/data
/web
/home/appl/data/web/important_big_spreadsheet.xls
/home/appl/data/web/big_architecture_drawing.ppt
/home/appl/data/web/unstructured_big_video.mpg
Policy EngineGlobal Namespace
Lo
gic
al P
hysic
al
✓Right Data
✓Right Place
✓Right Time
✓Right Performance
✓Right Cost
Tier 1: SSD/Flash Tier 2: SAS drives 15k Tier 3: 14TB SAS-NL drives Tier 4: Tape 30 TB
Business Challenge
To maintain its reputation
as a premier research
institution, the University
of Birmingham must
ensure data is always
available to a growing
number of users running
increasingly complex
simulations
Supports
Compliance with data
protection at low cost
and without disruption
Up to 2 FTEs
Estimated savings due
to enhanced operational
efficiency
“Spectrum Scale gives us unprecedented
insight into who is using data and how,
which we can use to meet reporting
requirements and ensure correct processes
are followed.”
Simon Thompson, Research Computing Infrastructure Architect
University of BirminghamDriving innovative research forward by taking control of data
Geography: Europe
Industry: Healthcare
IBM Solution Components:
IBM Spectrum Scale ™, IBM Spectrum Protect ™
Case Study
Video
Business Challenge
Posti Messaging’s
clients expect its
systems to be available
24/7, and its service
level agreements (SLAs)
demand recovery times
of less than 30 minutes.
How could the company
meet these challenging
requirements?
30-minute
Recovery time enables to
meet SLAs
2x faster
storage performance, with
synchronous replication
40% reduction
In storage requirements
enabled by built-in
compression capabilities
“The Elastic Storage Servers are noticeably
faster than our other storage systems in the
private cloud. That performance boost gives
us the speed we will need for new projects
such as machine learning and artificial
intelligence initiatives.”
Majid Ali, Head of Business IT
PostiKeeping Finland’s transactional, documents and messaging services on time with a robust storage cluster
Geography: Europe
Industry: Travel & Transportation
IBM Solution Components:
IBM Elastic Storage Server, IBM Spectrum Scale ™,
IBM Spectrum Protect ™
Case Study
Spectrum Protect
Server
Backup to Disk to Tape mit Spectrum Protect ( TSM )
21
ESS
filesystem
Spectrum Scale
IBM Spectrum Scale 2x TSM server
• Peak performance per TSM server 5.4 GB/s
• Peak performance total 9 GB/s
• Backup data and TSM db/log on IBM ESS:
no additional storage system needed!
• Easy integration of tape storage (reducing TCO)
Spectrum Protect
Server
Tape
Library
Spectrum Scale
Spectrum Scale
alte Werte
http://www.theedison.com/pdf/2015_Samples_IBM_Spectrum_Scale_WP.pdf
IBM Systems
Cluster A ‚Europe‘
Cluster B ‚US‘
/gpfs1_clusterA
/gpfs2_clusterB
LAN / WANvia TCP/IP
Cluster C ‚Far East Asia‘
Kunde sucht global verteilten Speicher - Niederlassungskonzepte
22
Cache
GPFSCache
GPFS
Cache
GPFS
IBM Systems
Asynchronous Disaster Recovery
▪ Asynchronously Replicate data from primary to secondary site using AFM
• Failover to secondary when Primary fails
• Failback when primary comes back
• Allow primary to operate actively with no interruption when the relationship with secondary fails
• Active / Passive model – Primary is Active (Read/Write) and secondary is passive (Read Only)
• Supports Recovery Point Objective (RPO)
• Recovery Time Objective (RTO) determined by network bandwidth,amount of data changed, number of files
• Usability: simpler and fewer commands to do multisite management
AFM Primary Site
Germany
AFM Secondary Site
Switzerland
Push all updates
asynchronously
App
Spectrum Scale
IBM Systems24
Compute nodes
Storage nodes
Kunde suchtHigh Performance – in Scientific Computing
Spectrum Scale
Spectrum Scale
Spectrum Scale
Spectrum Scale
Spectrum Scale
Spectrum Scale
Spectrum Scale
Spectrum Scale
25
CORAL ORNL (Summit) 200PF System – 250PB Usable Storage
POWER9 2 Socket Server2 P9 + 6 Volta GPU
512 GiB SMP Memory (32 GiB DDR4 RDIMMs)
96 GiB GPU Memory (HBM stacks)
1.6 TB NVMe
Compute Rack:
18 nodes
775 TF/s
10.7 TiB
59 KW max
Mellanox
IB4X EDR Switch IB-2
Floor plan rack conceptCompute (256)Switch (18)Storage (40)Infrastructure (4)
Standard 2U 19in. Rack
mount Chassis
POWER9:22 Cores4 Threads/core0.54 DP TF/s3.07 GHz
ESS Building
Block
SXM2
Volta:7.0 DP TF/s16GB @ 0.9 TB/s
SSC (4 ESS GL4):
8 servers, 16 JBOD
16.8 PB (gross)
38 KW max
Mellanox
IB4X EDR
648p Directors
Full bisection
25
Business Challenge
To advance our
understanding of the
human genome,
scientists must process
vast amounts of data.
However, many research
centers struggle to cope
with volume of data they
generate.
96%
reduction in the runtime of a
standard genome analysis
pipeline
1/3
The price of using commodity
solutions to perform the same
work at scale
2 weeks
From conceptual design to fully
functional IBM HPC environment
in the cloud
“IBM Spectrum Scale provides high-
performance data storage that we can scale
quickly and easily. Built-in tiering capabilities
allow a lot of flexibility in how we move data
around, enabling customers to seamlessly
migrate data from lab instruments up to the
cloud for analysis and long-term storage.“
Chris Mueller, Founder
L7 InformaticsHigh-performance Genomic Cloud for ground-breaking research
Geography: NA
Industry: Computer Services
IBM Solution Components:
IBM Spectrum Scale ™
Case Study
Video
By scheduling workloads intelligently according to policy, IBM Spectrum LSF reduces application run-time as well as prioritizes and optimizes resource use.
The business value of IBM Spectrum LSF
VIRTUALIZED POOL OF COMPUTE, NETWORK AND STORAGE RESOURCES
FASTER MORE
EFFICIENT
Spectrum Conductor
Spectrum LSF
Spectrum Symphonie
IBM Systems
Reference DESY
Kunde suchtHigh Performance – in Scientific Computing
Kunde sucht
High Performance – in Videoüberwachung• Intelligent Video Analysis improves Airport Safety and Efficiency
• With AI technology, machine could understand each video frame and allow Airport
authorities to respond to real-time events
Trolley Counting
Management Sys
Left Passenger
Object on APM
Face Recognition &
People Tracking
Person identification
Unattended Object
Detection
Intrusion Detection
Crowd Detection &
Queue Mgmt
Retail Analytics
Aircraft Parking
Clearance
Tailgating
License Plate
Recognition
Vehicle Counting
4k Kameras
Automatic Labeling using PowerAI Vision
31
Train DL ModelManually Label Some Image / Video Frames
Auto-Label Full Dataset with Trained DL Model
Manually Correct Labels on Some Data
Repeat Till Labels Achieve Desired Accuracy
32
33
IBM Storage & SDI
© Copyright IBM Corporation 2018
Mellanox IB Networking
NVIDIA DGX Servers
IBM NVMe_Powered_ESS 3000Densest and fastest storage
> 90 GB/s throughput
Kunde sucht
NVMe als High Performance Storage Tier
35
IBM Systems
Kunde sucht Analytics Lösung
SAS Viya
37https://apps.na.collabserv.com/communities/service/html/communityview?communityUuid=497eb8e7-55e1-4b87-bc59-
IBM Systems
Kunde suchtAnalytics Lösung DWH Daten ins Hadoop
Schadens-
daten
Vertrags-
daten
Kunden-
daten
.
.
.
DWHETL-Prozesse
Fach-
bereich
Fach-
bereich
Fach-
bereich
Vorteile der Lösung:
- Ältere Daten werden auf eine
kostengünstigere Infrastruktur
ausgelagert
- Flexibler Zugriff auf Hadoop und
DWH Daten durch BigSQL
(Federation Layer)
- Zugriffsrechte bleiben über die
BigSQL Sicht erhalten
- BigSQL als Basis für übergreifende
Analytics Anwendungen
2001
2002
2003
H
a
d
o
o
p
Big SQL
Entladen
DSX
Spectrum Scale - enhancing HDFS Big Data deployments
26
Full tech details at IBM developerWorks:
http://ibm.co/2m2UR9B
Data × 3
virtual (network-shared) disks
Data × 1.3
replicating disks
classic HDFS-based on IBM Elastic Storage
Eth
Shared disk
server (ESS)
with Spectrum
Scale
Business Challenge
What causes some
people to develop
diseases and not
others? The attempt to
find an answer is driving
groundbreaking
research and leading
pioneers to challenge
traditional approaches to
treatment.
Pushes
the boundaries of
knowledge, anticipating new
breakthroughs in healthcare
Supports
the development of
diagnostic and therapies
Removes
barriers to scientific
exploration through data-
driven research
“In a relatively short time, we recovered
years’ worth of data, which had been
generated by dozens of people in the Center.
And, before long, we were again able to shift
our attention back to our research work.“
Dr. Isidore Rigoutsos, Director of the Computational Medicine Center
Jefferson Optimizing data-driven research
Geography: NA
Industry: Healthcare
IBM Solution Components:
IBM Spectrum Scale ™, IBM Spectrum Protect ™,
IBM Storwize ® V5030, IBM TS3310 Tape Library
Case Study
Videos
Business Challenge
To deliver life-changing
care, UPMC must
ensure clinical and
research teams have
fast, reliable access to
data. How could the
organization shorten
storage response times
and increase
availability?
2,400%
growth in storage under
management, supported
with zero increase in IT
headcount
50%
TCO saving for primary
storage
50%
shorter patient record seek
times
“On the hospital floor, IBM Storage solutions
are having a real impact: orders go in faster,
test results are available sooner, and uptime
is improved across all systems.“
Kevin Muha, Director of Storage and Data Protection
UPMCEnabling quality patient care and improving user experience by delivering rapid access to vital data
Geography: NA
Industry: Healthcare
IBM Solution Components:
IBM Spectrum Scale™, IBM Spectrum Protect™,
IBM Spectrum Accelerate™, SVC, IBM
FlashSystem ® 900, IBM FlashSystem ® A9000
Case Study
© 2019 IBM Corporation
IBM Systems
Kunde suchtObjekt Speicher
Projekt
http://iot-cosmos.eu/
IBM Systems
Use case – Enabling “In-Place” analytics for Object data repository
Spectrum Scale
<SOF_Fileset>/<Device>
Object
(http)
Data ingested
as Objects
Spectrum Scale
Hadoop Connectors
In-Place Analytics
Source:https://aws.amazon.com/elasticmapreduce/
Traditional object store – Data to be copied from
object store to dedicated cluster , do the analysis
and copy the result back to object store for
publishing
Spectrum Scale object store with Unified File and Object Access –
Object Data available as File on the same fileset . Spectrum Scale Hadoop
connectors allow the data to be directly leveraged for analytics.
No data movement / In-Place immediate data analytics.
Analytics on Spectrum Scale Object Store With Unified
File and Object AccessAnalytics on Traditional Object Store
Explicit Data movement
Results Published
as Objects with
no data movement
Results returned
in place
IBM Systems
Kunde suchtEnterprise Synch and Share für Smart Devices OnPremise
CC
CC
CC
GPFS
ESS ESS ESS
Sync & Share
Use Case Versicherung
1. Sharen von Daten für den Aufsichtsrat
2. Versicherung zu Makler Kommunikation
3. Externe Kommunikation mit Anwälten
Team File Sharing
ohne Access Rights
MicroManagement
IBM Systems
Hardware components:• FlashSystem 9100• Cloud Object Storage• Z LinuxONE Rockhopper II• Networking switches
Options
Expansion Options
• Performance Optimized• Capacity Optimized
A full Multicloud Storage Solutions for Blockchain stack contains:
• Hardware components
• Software components
• Ordering and Installation
• Expansion options
The Solution Blueprint will tie the whole solution together with a set of instructions defining each component and the required setup
Installation SMEs available to help you through your installation and setup journey
Payment Options
• Purchase or Utility pricing• Subscription License Model
Solution Blueprint with setup instructions
Software components:• Spectrum Scale• Spectrum Virtualize• Spectrum CDM• Spectrum Connect / Storage
Enabler for containers
Platforms:• Distributed/Remote Peer• IBM Cloud Private
Support:• Pre & Post Sales support
• Integrated Support Team
“A secure end to
end solution easy
to maintain”
“Predefined solution
with setup
instructions”
Kunde suchtBlockchain Lösung für 'Value content'
IBM Systems
Architectural Diagram
Blockchain Network
Built on IBM Hyperledger Fabric
Off-Chain
Node
Networking
Compute
Blockchain networks run on a
set of distributed nodes or
peers.
Each node holds a copy of
the shared ledger.
Participants
Peers act as a gateway into
the network
Participants must interact
with the blockchain network
through a peerDistributed Peer
On IBM Cloud Private
Participant
Participant
Participants Participants Participants
Offchain
Extension
Offchain
Extension
Participant
Off-Chain
Node
Networking
Compute
Cloud Peer
Participant Participant Participant
On-Chain
On-Chain
IBM Systems
Danke