SlideShare a Scribd company logo
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
Data Volumes and I/O Pressures are Compounding
Quantum
Monte Carlo
LHC
Plasma Physics
SKA
LSST
Nanopore
100+TB/s
200TB/day
100TB/s
1GB/s
CCTV
AD
data collection
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
AI-HPC
Simulation Big Data Deep Learning
Materials Science APS Data Analysis Drug Response Prediction
Cosmology HEP Data Analysis Scientific Image
Classification
Molecular Dynamics LSST Data Analysis Scientific Text
Understanding
Nuclear Reaction
Modelling
SKA Data Analytics Materials Property Design
Combustion Metagenome
Analysis
Gravitational Lens
Detection
Quantum Computer
Simulation
Graph analysis Feature Detection in 3D
Climate Modelling Virtual compound
library
Street Scene Analysis
Power Grid Neuroscience data
analysis
Organism Design
Discrete Event
Simulation
Genome Pipelines State Space Prediction
Fusion Reactor
Simulation
Persistent Learning
Transportation
Networks
Hyperspectral patterns
Traditional
HPC
Large Scale
Numerical
Simulation
Scalable Data
Analytics
Deep Learning
DDN ©2019 DataDirect Networks, Inc.
XXII Computer Vision and AI for Smart Retail
Flow management. People counting. Product counting.
Autonomous check-out. Staff access control.
One system, endless data.
► Intelligence built on a full network of cameras
► Object identification and classification in real-time
► Skeletal and facial human recognition
► One system, multiple uses, in real-time:
• Store Analytics – flow management to optimize resources
• Frictionless Shopping – seamless in-store experience to allow staff to
focus on customers
• People Safety – create safer environments and decrease human risk
• Supply Chain Control – enhance quality and productivity
Item 07482
DDN ©2019 DataDirect Networks, Inc.
ToMMo Architecture for Genomics Acceleration
Tokyo Medical Megabank Organization
Accelerating genomic medicine research from a more robust Biobank by
providing wider access and better performance to an expanded storage
infrastructure.
► Shared computational resources made up of a 300 compute node
cluster, 3 shared memory compute nodes, and 3 NVIDIA DGX-1 systems
► Mixed Infiniband and 40/10 Gbps networks to provide capabilities for
different needs
► Analysis system (16PB)
► Public data system (7PB EXA)
► 2 x Information distribution systems (4PB EXA, 2PB GRID)
► DGX-1 systems are using EXA to run Parabricks genomics analysis at a
much accelerated rate
DDN is expected to continue to proactively propose solutions leveraging the
newest technology to accelerate research. Compared to when the system
was initially implemented, the organization carries heavier burden of social
responsibility. Though the system’s original intended use was for analysis, it
has become a system upon which many researchers have built their projects.
GRID
16PB Analysis 7PB Public Data 6PB Info
Distribution
Diverse Compute Resources
CPU GPU
DDN ©2019 DataDirect Networks, Inc.
TACC Architecture and NVMe acceleration
► The primary computing system (Dell/EMC, Intel,
Mellanox HDR) focussed on high precision
performance
► Initial configuration of the system will have 8,008
available compute nodes.
► second subsystem focused on single precision
streaming-memory computing.
► multiple storage systems
► interfaces to cloud and archive systems
► application nodes for hosting virtual servers
Scratch/Work
Disk based Filesystem
50PB, 240GB/s
NVMe Acceleration
3PB, 1.2TB/s
Capability: LNET routers, fs scanning,
user and fs monitoring, project
quota, data management, security,
end-to-end data protection
Mixed double precision cpu and single precision compute
computational capability that makes it possible for investigators to
tackle much larger and more complex research challenges across a wide
spectrum of domains.
DDN ©2019 DataDirect Networks, Inc.
So where are the Pressures?
► Data Volumes going up (HPC and IA)
► Flash Pricing Going down.. Still some market instabilities, But HDD prices going down too!
► Performance requirements going up (of course)
• Ingest (AI (new scenarios) and HPC )
• Mixed IO (HPC and AI)
• Single client IO and mmap() IO (AI)
► Uptimes and User Management requirements up (HPC and AI)
► Feature Set going up, Access and Protocols (primary driver from AI)
► Protocol access requirements going up (AI)
► Security going up - Encryption, External KEY Mgmt (general trend)
► MultiCloud demand rising (mainly AI, some HPC)
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
EXA5 Capability for the AI-HPC DataCenter
EXTREME EDGE
AI
Move, ManageImport, Preprocess Tune, DeployInfer, PredictClassify, Tag Train, Score
Management, Oversight, Governance
HA, Uptime Backup, Archive, Sync
Scalable Filesystem
Security
NFS CIFS HDFS
GPU/Container Integration
DATA MANAGEMENT
Workflows Grand Challenge Big Data Multi-Physics
HPC
Mesh RefinementNoSQL
INFRASTRUCTURE
GPU CPU
GPUs & Novel Processors
CPU
Broader Range of Credible CPUsEDR, HDR, Ethernet and More!
IPU
Cloud, Hybrid Cloud, MultiCloud
User & Service Management
More access protocols
Strong Data Protection
Comprehensive Security
Full Spectrum Performance
User and Workload management
Complete Data Management
Multiple Cloud options
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
POSIX
NFS
SMB
HDFS
S3
NVMe, SSD, HDD, TAPE, S3
EXA
5
Comprehensive Security
Full Spectrum Performance
Complete Data Management
Innovation at Scale
Google Cloud and DDN have developed an
in-cloud file system suitable for HPC: DDN’s
EXAScaler, a parallel file system designed to
handle high concurrency access patterns to
shared data sets.
*
Strong Data Protection
User & Workload Management
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
EXA5 optimizations for AI Workloads
► DDN AI200 with 20 x 1.6TB PCI NVMe
devices
• 32x clients, 2x 12 CPU cores, 128GB memory
► EXA5 37% higher 4K IOPS than 2.10.7
• 70% of peak IOPS efficiency vs. RAM-only workload
► EXA5 is better than 2.10.7 at every IO size
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
COMPREHENSIVE FLASH MANAGEMENT
► Complete flash management requires
integration with the Flash capability to
manage garbage collection cycles
► EXA5 integrations with DDN’s SFA
architecture to deliver unmap operations
from the filesystems allowing total control
over NAND Flash Devices
► Manage Flash Performance Behavior with
EXA5 FSTRIM
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
Managing AI Workloads
EXA5 MASS STORAGE
INGEST
Stream Engine
Spark
CLASSIFICATION MODEL DEVELOPMENT
Ingest Clean Train Validate Infer
Access Pattern Sequential
Sequential or
Random
Random Sequential Sequential
Access Type Write Read / Write Read Read Read
Concurrency
Varies by number of
sources
High High High
Varies by use
case
Size of File
Metadata small; Files
vary by source
Vary by source
Small access
within large
files
Varies
Varies by use
case
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
Comprehensive User & Workload Monitoring
► Out of the box access to your filesystem or user’s
file distribution profile
► How many files, size, statistics
► Real-time access to granular properties of live
workloads (jobstats)
► Simpler diagnostics of client issues and networking
challenges.
► Easy identification of problem jobs (high metadata,
dominating throughput, etc)
File Size Distribution
User 1 Metadata
User 1 Throughput
FileCountOpsGB/s
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
All Round Security, Multi-Tenancy, Audit and Encryption
Group A Group C
► Managing Multiple Tenants in complex security
environments
► Partition your users strictly at a sub filesystem
► Full FIPs/KMIP support for formal certification
of encryption
► Access Control
► Audit Facility - records of all EXAScaler accesses
using efficient Lustre changelog
• Record file system namespace & metadata events in
secure log
• Record Failed access attempts
Group B
AUDIT
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
EXA5 NFS and CIFS Data Services
Consistency and locking semantics with Linux/Unix and Windows
clients
Export Lustre namespace through CIFS and NFS gateway
Authentication methods for UNIX passwords, LDAP and Microsoft
AD
High Availability
Horizontal scalability
• Formal, Qualified and Tested NAS
Solutions for DDN EXAScaler
• Performance Class NFS and CIFS for
scalable Filesystem export
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
EXA5 END TO END DATA PROTECTION
End-to-EndProtection
APPLICATION/USERSPACE
EXA5 CLIENT
NETWORK
EXA5 SERVER
Integrated, end-to-end data protection using T10PI/DIX
Standard
Fully Transparent to Applications
Supported by EXAScaler from the client through to Drives
Minimal Performance Impact
► Integrated, end-to-end data protection
using T10PI/DIX Standard
► Fully Transparent to Applications
► Supported by EXAScaler from the
client through to Drives
► Minimal Performance Impact
T10-PI
DDN ©2019 DataDirect Networks, Inc.
►Fast and efficient data management, built-in.
►Manage your most active data into scale-out Flash
►Automatic control of free space on Flash
►Quickly respond to changing demand
►Efficient namespace scanning
►Built in To EXA5 - No external servers
EXA5 Native Scanning Engine
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
Hybrid at Scale
Mixed I/O
Mixed file sizes
Stratagem
manages Hot data
automatically
tiering between
flash and HDD
Small file data
automatically goes into
metadata
 Fewer RPCs
 Lower latencies
► Flexible Small File Acceleration
brings Simple, Powerful
Application of Flash and Flash
optimized IO to Scalable Hybrid
Storage Systems
► Expand the domain of file sizes
accelerated simply from a few
KB to Multiple MB
ScaleOutScaleOut
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
POSIX
NFS
SMB
HDFS
S3
NVMe, SSD, HDD, TAPE, S3
EXA
5
Comprehensive Security
Full Spectrum Performance
Complete Data Management
Innovation at Scale
Google Cloud and DDN have developed an
in-cloud file system suitable for HPC: DDN’s
EXAScaler, a parallel file system designed to
handle high concurrency access patterns to
shared data sets.
*
• Data on Metadata
• File Level Redundancy
• Hot Pools
• Scale Out Metadata
• NVMe Optimizations
• Stratagem Scanner
• Automatic Tiering
• DataFlow Integration
• Multi-Tenancy
• Audit
• Kerberos
• Access Control
Strong Data Protection
• DeClustered RAID
• End-to-End data integrity
• Partial Rebuilds
• Enclosure Redundancy
DDN Confidential
DDN ©2019 DataDirect Networks, Inc.
Innovation, Stability,
Partnership

More Related Content

PDF
Optimizing Lustre and GPFS with DDN
PDF
DDN: Massively-Scalable Platforms and Solutions Engineered for the Big Data a...
PDF
HPC at Scale Enabled by DDN A3i and NVIDIA SuperPOD
PPTX
NetApp Se training storage grid webscale technical overview
PPTX
Present of Raid and Its Type
PDF
MapReduce
PPSX
HPE SimpliVity
PDF
NVMe overview
Optimizing Lustre and GPFS with DDN
DDN: Massively-Scalable Platforms and Solutions Engineered for the Big Data a...
HPC at Scale Enabled by DDN A3i and NVIDIA SuperPOD
NetApp Se training storage grid webscale technical overview
Present of Raid and Its Type
MapReduce
HPE SimpliVity
NVMe overview

What's hot (20)

PPTX
Introduction to Unix
PDF
An introduction to the linux kernel and device drivers (NTU CSIE 2016.03)
PDF
DDN: Protecting Your Data, Protecting Your Hardware
PPTX
Windows Operating system
PDF
Cyclone DDS: Sharing Data in the IoT Age
PPTX
Q1 Memory Fabric Forum: Compute Express Link (CXL) 3.1 Update
PPTX
Ibm spectrum scale_backup_n_archive_v03_ash
ODP
Software defined storage
PDF
DAIS19: On the Performance of ARM TrustZone
PPTX
Light-weighted HDFS disaster recovery
PPTX
AI Hardware Landscape 2021
PDF
Linux Introduction
PDF
Software-Defined Storage (SDS)
PDF
Hands-on Lab: How to Unleash Your Storage Performance by Using NVM Express™ B...
PDF
Disk allocation methods
PPTX
Linux.ppt
PDF
Scaling Databricks to Run Data and ML Workloads on Millions of VMs
PDF
Linux Kernel and Driver Development Training
PDF
Scheduling in Android
PPTX
Linux operating system - Overview
Introduction to Unix
An introduction to the linux kernel and device drivers (NTU CSIE 2016.03)
DDN: Protecting Your Data, Protecting Your Hardware
Windows Operating system
Cyclone DDS: Sharing Data in the IoT Age
Q1 Memory Fabric Forum: Compute Express Link (CXL) 3.1 Update
Ibm spectrum scale_backup_n_archive_v03_ash
Software defined storage
DAIS19: On the Performance of ARM TrustZone
Light-weighted HDFS disaster recovery
AI Hardware Landscape 2021
Linux Introduction
Software-Defined Storage (SDS)
Hands-on Lab: How to Unleash Your Storage Performance by Using NVM Express™ B...
Disk allocation methods
Linux.ppt
Scaling Databricks to Run Data and ML Workloads on Millions of VMs
Linux Kernel and Driver Development Training
Scheduling in Android
Linux operating system - Overview
Ad

Similar to DDN EXA 5 - Innovation at Scale (20)

PDF
DDN and Intel: Partnered for Exascale
PDF
DDN Product Update from SC13
PPTX
Accelerated Any-Scale Solutions from DDN
PDF
DDN Strategic Vision Tour June 2015
PPTX
Innovating to Create a Brighter Future for AI, HPC, and Big Data
PDF
Long Live Posix - HPC Storage and the HPC Datacenter
PDF
DDN-DataFlow-Deck_v1.pdf
PDF
Ddn Vision
PDF
Big Data: Infrastructure Implications for “The Enterprise of Things” - Stampe...
PDF
DDN Accelerating-Decisions-Through-Enterprise-Hadoop-final
PPTX
Hype, Hopes, Hell & Hadoop (#bigdata and the enterprise of everything)
PPTX
Webinar: Getting Beyond Flash 101 - Flash 102 Selecting the Right Flash Array
PDF
IME - Unlocking the Potential of NVMe
PDF
IO Management with IME 1.1
PDF
A Glimpse into the Future of I/O
PDF
DELL EMC DEA-1TT5 Updated Dumps 2023
PDF
DDN Service Strategy
PDF
Infinite Memory Engine: HPC in the FLASH Era
PDF
prm4114-exadatastrategy.pdf
PDF
Building a High Performance Analytics Platform
DDN and Intel: Partnered for Exascale
DDN Product Update from SC13
Accelerated Any-Scale Solutions from DDN
DDN Strategic Vision Tour June 2015
Innovating to Create a Brighter Future for AI, HPC, and Big Data
Long Live Posix - HPC Storage and the HPC Datacenter
DDN-DataFlow-Deck_v1.pdf
Ddn Vision
Big Data: Infrastructure Implications for “The Enterprise of Things” - Stampe...
DDN Accelerating-Decisions-Through-Enterprise-Hadoop-final
Hype, Hopes, Hell & Hadoop (#bigdata and the enterprise of everything)
Webinar: Getting Beyond Flash 101 - Flash 102 Selecting the Right Flash Array
IME - Unlocking the Potential of NVMe
IO Management with IME 1.1
A Glimpse into the Future of I/O
DELL EMC DEA-1TT5 Updated Dumps 2023
DDN Service Strategy
Infinite Memory Engine: HPC in the FLASH Era
prm4114-exadatastrategy.pdf
Building a High Performance Analytics Platform
Ad

More from inside-BigData.com (20)

PDF
Major Market Shifts in IT
PDF
Preparing to program Aurora at Exascale - Early experiences and future direct...
PPTX
Transforming Private 5G Networks
PDF
The Incorporation of Machine Learning into Scientific Simulations at Lawrence...
PDF
How to Achieve High-Performance, Scalable and Distributed DNN Training on Mod...
PDF
Evolving Cyberinfrastructure, Democratizing Data, and Scaling AI to Catalyze ...
PDF
HPC Impact: EDA Telemetry Neural Networks
PDF
Biohybrid Robotic Jellyfish for Future Applications in Ocean Monitoring
PDF
Machine Learning for Weather Forecasts
PPTX
HPC AI Advisory Council Update
PDF
Fugaku Supercomputer joins fight against COVID-19
PDF
Energy Efficient Computing using Dynamic Tuning
PDF
State of ARM-based HPC
PDF
Versal Premium ACAP for Network and Cloud Acceleration
PDF
Zettar: Moving Massive Amounts of Data across Any Distance Efficiently
PDF
Scaling TCO in a Post Moore's Era
PDF
CUDA-Python and RAPIDS for blazing fast scientific computing
PDF
Introducing HPC with a Raspberry Pi Cluster
PDF
Overview of HPC Interconnects
PDF
Efficient Model Selection for Deep Neural Networks on Massively Parallel Proc...
Major Market Shifts in IT
Preparing to program Aurora at Exascale - Early experiences and future direct...
Transforming Private 5G Networks
The Incorporation of Machine Learning into Scientific Simulations at Lawrence...
How to Achieve High-Performance, Scalable and Distributed DNN Training on Mod...
Evolving Cyberinfrastructure, Democratizing Data, and Scaling AI to Catalyze ...
HPC Impact: EDA Telemetry Neural Networks
Biohybrid Robotic Jellyfish for Future Applications in Ocean Monitoring
Machine Learning for Weather Forecasts
HPC AI Advisory Council Update
Fugaku Supercomputer joins fight against COVID-19
Energy Efficient Computing using Dynamic Tuning
State of ARM-based HPC
Versal Premium ACAP for Network and Cloud Acceleration
Zettar: Moving Massive Amounts of Data across Any Distance Efficiently
Scaling TCO in a Post Moore's Era
CUDA-Python and RAPIDS for blazing fast scientific computing
Introducing HPC with a Raspberry Pi Cluster
Overview of HPC Interconnects
Efficient Model Selection for Deep Neural Networks on Massively Parallel Proc...

Recently uploaded (20)

PDF
Per capita expenditure prediction using model stacking based on satellite ima...
PDF
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
PDF
Reach Out and Touch Someone: Haptics and Empathic Computing
PDF
CIFDAQ's Market Insight: SEC Turns Pro Crypto
PDF
Encapsulation_ Review paper, used for researhc scholars
PDF
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
PDF
NewMind AI Weekly Chronicles - August'25 Week I
PDF
Modernizing your data center with Dell and AMD
PPTX
Big Data Technologies - Introduction.pptx
PDF
Electronic commerce courselecture one. Pdf
PPTX
PA Analog/Digital System: The Backbone of Modern Surveillance and Communication
PDF
Agricultural_Statistics_at_a_Glance_2022_0.pdf
PDF
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
PDF
Machine learning based COVID-19 study performance prediction
PDF
Dropbox Q2 2025 Financial Results & Investor Presentation
PDF
Review of recent advances in non-invasive hemoglobin estimation
PPTX
MYSQL Presentation for SQL database connectivity
DOCX
The AUB Centre for AI in Media Proposal.docx
PDF
Blue Purple Modern Animated Computer Science Presentation.pdf.pdf
PDF
Chapter 3 Spatial Domain Image Processing.pdf
Per capita expenditure prediction using model stacking based on satellite ima...
Peak of Data & AI Encore- AI for Metadata and Smarter Workflows
Reach Out and Touch Someone: Haptics and Empathic Computing
CIFDAQ's Market Insight: SEC Turns Pro Crypto
Encapsulation_ Review paper, used for researhc scholars
Build a system with the filesystem maintained by OSTree @ COSCUP 2025
NewMind AI Weekly Chronicles - August'25 Week I
Modernizing your data center with Dell and AMD
Big Data Technologies - Introduction.pptx
Electronic commerce courselecture one. Pdf
PA Analog/Digital System: The Backbone of Modern Surveillance and Communication
Agricultural_Statistics_at_a_Glance_2022_0.pdf
How UI/UX Design Impacts User Retention in Mobile Apps.pdf
Machine learning based COVID-19 study performance prediction
Dropbox Q2 2025 Financial Results & Investor Presentation
Review of recent advances in non-invasive hemoglobin estimation
MYSQL Presentation for SQL database connectivity
The AUB Centre for AI in Media Proposal.docx
Blue Purple Modern Animated Computer Science Presentation.pdf.pdf
Chapter 3 Spatial Domain Image Processing.pdf

DDN EXA 5 - Innovation at Scale

  • 1. DDN Confidential DDN ©2019 DataDirect Networks, Inc.
  • 2. DDN Confidential DDN ©2019 DataDirect Networks, Inc. Data Volumes and I/O Pressures are Compounding Quantum Monte Carlo LHC Plasma Physics SKA LSST Nanopore 100+TB/s 200TB/day 100TB/s 1GB/s CCTV AD data collection
  • 3. DDN Confidential DDN ©2019 DataDirect Networks, Inc. AI-HPC Simulation Big Data Deep Learning Materials Science APS Data Analysis Drug Response Prediction Cosmology HEP Data Analysis Scientific Image Classification Molecular Dynamics LSST Data Analysis Scientific Text Understanding Nuclear Reaction Modelling SKA Data Analytics Materials Property Design Combustion Metagenome Analysis Gravitational Lens Detection Quantum Computer Simulation Graph analysis Feature Detection in 3D Climate Modelling Virtual compound library Street Scene Analysis Power Grid Neuroscience data analysis Organism Design Discrete Event Simulation Genome Pipelines State Space Prediction Fusion Reactor Simulation Persistent Learning Transportation Networks Hyperspectral patterns Traditional HPC Large Scale Numerical Simulation Scalable Data Analytics Deep Learning
  • 4. DDN ©2019 DataDirect Networks, Inc. XXII Computer Vision and AI for Smart Retail Flow management. People counting. Product counting. Autonomous check-out. Staff access control. One system, endless data. ► Intelligence built on a full network of cameras ► Object identification and classification in real-time ► Skeletal and facial human recognition ► One system, multiple uses, in real-time: • Store Analytics – flow management to optimize resources • Frictionless Shopping – seamless in-store experience to allow staff to focus on customers • People Safety – create safer environments and decrease human risk • Supply Chain Control – enhance quality and productivity Item 07482
  • 5. DDN ©2019 DataDirect Networks, Inc. ToMMo Architecture for Genomics Acceleration Tokyo Medical Megabank Organization Accelerating genomic medicine research from a more robust Biobank by providing wider access and better performance to an expanded storage infrastructure. ► Shared computational resources made up of a 300 compute node cluster, 3 shared memory compute nodes, and 3 NVIDIA DGX-1 systems ► Mixed Infiniband and 40/10 Gbps networks to provide capabilities for different needs ► Analysis system (16PB) ► Public data system (7PB EXA) ► 2 x Information distribution systems (4PB EXA, 2PB GRID) ► DGX-1 systems are using EXA to run Parabricks genomics analysis at a much accelerated rate DDN is expected to continue to proactively propose solutions leveraging the newest technology to accelerate research. Compared to when the system was initially implemented, the organization carries heavier burden of social responsibility. Though the system’s original intended use was for analysis, it has become a system upon which many researchers have built their projects. GRID 16PB Analysis 7PB Public Data 6PB Info Distribution Diverse Compute Resources CPU GPU
  • 6. DDN ©2019 DataDirect Networks, Inc. TACC Architecture and NVMe acceleration ► The primary computing system (Dell/EMC, Intel, Mellanox HDR) focussed on high precision performance ► Initial configuration of the system will have 8,008 available compute nodes. ► second subsystem focused on single precision streaming-memory computing. ► multiple storage systems ► interfaces to cloud and archive systems ► application nodes for hosting virtual servers Scratch/Work Disk based Filesystem 50PB, 240GB/s NVMe Acceleration 3PB, 1.2TB/s Capability: LNET routers, fs scanning, user and fs monitoring, project quota, data management, security, end-to-end data protection Mixed double precision cpu and single precision compute computational capability that makes it possible for investigators to tackle much larger and more complex research challenges across a wide spectrum of domains.
  • 7. DDN ©2019 DataDirect Networks, Inc. So where are the Pressures? ► Data Volumes going up (HPC and IA) ► Flash Pricing Going down.. Still some market instabilities, But HDD prices going down too! ► Performance requirements going up (of course) • Ingest (AI (new scenarios) and HPC ) • Mixed IO (HPC and AI) • Single client IO and mmap() IO (AI) ► Uptimes and User Management requirements up (HPC and AI) ► Feature Set going up, Access and Protocols (primary driver from AI) ► Protocol access requirements going up (AI) ► Security going up - Encryption, External KEY Mgmt (general trend) ► MultiCloud demand rising (mainly AI, some HPC)
  • 8. DDN Confidential DDN ©2019 DataDirect Networks, Inc. EXA5 Capability for the AI-HPC DataCenter EXTREME EDGE AI Move, ManageImport, Preprocess Tune, DeployInfer, PredictClassify, Tag Train, Score Management, Oversight, Governance HA, Uptime Backup, Archive, Sync Scalable Filesystem Security NFS CIFS HDFS GPU/Container Integration DATA MANAGEMENT Workflows Grand Challenge Big Data Multi-Physics HPC Mesh RefinementNoSQL INFRASTRUCTURE GPU CPU GPUs & Novel Processors CPU Broader Range of Credible CPUsEDR, HDR, Ethernet and More! IPU Cloud, Hybrid Cloud, MultiCloud User & Service Management More access protocols Strong Data Protection Comprehensive Security Full Spectrum Performance User and Workload management Complete Data Management Multiple Cloud options
  • 9. DDN Confidential DDN ©2019 DataDirect Networks, Inc. POSIX NFS SMB HDFS S3 NVMe, SSD, HDD, TAPE, S3 EXA 5 Comprehensive Security Full Spectrum Performance Complete Data Management Innovation at Scale Google Cloud and DDN have developed an in-cloud file system suitable for HPC: DDN’s EXAScaler, a parallel file system designed to handle high concurrency access patterns to shared data sets. * Strong Data Protection User & Workload Management
  • 10. DDN Confidential DDN ©2019 DataDirect Networks, Inc. EXA5 optimizations for AI Workloads ► DDN AI200 with 20 x 1.6TB PCI NVMe devices • 32x clients, 2x 12 CPU cores, 128GB memory ► EXA5 37% higher 4K IOPS than 2.10.7 • 70% of peak IOPS efficiency vs. RAM-only workload ► EXA5 is better than 2.10.7 at every IO size
  • 11. DDN Confidential DDN ©2019 DataDirect Networks, Inc. COMPREHENSIVE FLASH MANAGEMENT ► Complete flash management requires integration with the Flash capability to manage garbage collection cycles ► EXA5 integrations with DDN’s SFA architecture to deliver unmap operations from the filesystems allowing total control over NAND Flash Devices ► Manage Flash Performance Behavior with EXA5 FSTRIM
  • 12. DDN Confidential DDN ©2019 DataDirect Networks, Inc. Managing AI Workloads EXA5 MASS STORAGE INGEST Stream Engine Spark CLASSIFICATION MODEL DEVELOPMENT Ingest Clean Train Validate Infer Access Pattern Sequential Sequential or Random Random Sequential Sequential Access Type Write Read / Write Read Read Read Concurrency Varies by number of sources High High High Varies by use case Size of File Metadata small; Files vary by source Vary by source Small access within large files Varies Varies by use case
  • 13. DDN Confidential DDN ©2019 DataDirect Networks, Inc. Comprehensive User & Workload Monitoring ► Out of the box access to your filesystem or user’s file distribution profile ► How many files, size, statistics ► Real-time access to granular properties of live workloads (jobstats) ► Simpler diagnostics of client issues and networking challenges. ► Easy identification of problem jobs (high metadata, dominating throughput, etc) File Size Distribution User 1 Metadata User 1 Throughput FileCountOpsGB/s
  • 14. DDN Confidential DDN ©2019 DataDirect Networks, Inc. All Round Security, Multi-Tenancy, Audit and Encryption Group A Group C ► Managing Multiple Tenants in complex security environments ► Partition your users strictly at a sub filesystem ► Full FIPs/KMIP support for formal certification of encryption ► Access Control ► Audit Facility - records of all EXAScaler accesses using efficient Lustre changelog • Record file system namespace & metadata events in secure log • Record Failed access attempts Group B AUDIT
  • 15. DDN Confidential DDN ©2019 DataDirect Networks, Inc. EXA5 NFS and CIFS Data Services Consistency and locking semantics with Linux/Unix and Windows clients Export Lustre namespace through CIFS and NFS gateway Authentication methods for UNIX passwords, LDAP and Microsoft AD High Availability Horizontal scalability • Formal, Qualified and Tested NAS Solutions for DDN EXAScaler • Performance Class NFS and CIFS for scalable Filesystem export
  • 16. DDN Confidential DDN ©2019 DataDirect Networks, Inc. EXA5 END TO END DATA PROTECTION End-to-EndProtection APPLICATION/USERSPACE EXA5 CLIENT NETWORK EXA5 SERVER Integrated, end-to-end data protection using T10PI/DIX Standard Fully Transparent to Applications Supported by EXAScaler from the client through to Drives Minimal Performance Impact ► Integrated, end-to-end data protection using T10PI/DIX Standard ► Fully Transparent to Applications ► Supported by EXAScaler from the client through to Drives ► Minimal Performance Impact T10-PI
  • 17. DDN ©2019 DataDirect Networks, Inc. ►Fast and efficient data management, built-in. ►Manage your most active data into scale-out Flash ►Automatic control of free space on Flash ►Quickly respond to changing demand ►Efficient namespace scanning ►Built in To EXA5 - No external servers EXA5 Native Scanning Engine
  • 18. DDN Confidential DDN ©2019 DataDirect Networks, Inc. Hybrid at Scale Mixed I/O Mixed file sizes Stratagem manages Hot data automatically tiering between flash and HDD Small file data automatically goes into metadata  Fewer RPCs  Lower latencies ► Flexible Small File Acceleration brings Simple, Powerful Application of Flash and Flash optimized IO to Scalable Hybrid Storage Systems ► Expand the domain of file sizes accelerated simply from a few KB to Multiple MB ScaleOutScaleOut
  • 19. DDN Confidential DDN ©2019 DataDirect Networks, Inc. POSIX NFS SMB HDFS S3 NVMe, SSD, HDD, TAPE, S3 EXA 5 Comprehensive Security Full Spectrum Performance Complete Data Management Innovation at Scale Google Cloud and DDN have developed an in-cloud file system suitable for HPC: DDN’s EXAScaler, a parallel file system designed to handle high concurrency access patterns to shared data sets. * • Data on Metadata • File Level Redundancy • Hot Pools • Scale Out Metadata • NVMe Optimizations • Stratagem Scanner • Automatic Tiering • DataFlow Integration • Multi-Tenancy • Audit • Kerberos • Access Control Strong Data Protection • DeClustered RAID • End-to-End data integrity • Partial Rebuilds • Enclosure Redundancy
  • 20. DDN Confidential DDN ©2019 DataDirect Networks, Inc. Innovation, Stability, Partnership

Editor's Notes

  • #18: DDN Enables and Accelerates Deep Learning DL is about growing autonomous capability Learning from a very large amounts of data In many ways, data is new source code Collect and access Success of DL development is access to large data sets The size, diversity and quality of data set critical Easy data discovery Store data in single location, accessible everywhere Presented unified interface, interoperable, standard No modification to application or container