Will Your Backup
Architecture Meet
Tomorrow’s SLAs?
3 Steps to Make Sure!
For audio playback and Q&A go to:
bit.ly/ExtremePerform
George Crump,
Lead Analyst & Founder
Thad Omura,
EVP Marketing &
Operations
Scott Shadley,
VP of Marketing
Pankaj Mehra,
VP, Product
Planning Team
Nick Adams,
Board Member
what is
COMPUTATIONAL
STORAGE?
why does it
MATTER?
what are some USE
CASES & SUCCESS
STORIES?
Presented by Scott Shadley, VP of Marketing
ngdsystems.com info@ngdsystems.com @NGDsystems
Form Factor
Available
In 2019
Capacity
(TB)
MAX Power
(W)
M.2 22110 NOW up to 8 8
EDSFF E1.S Q3 up to 16 12
EDSFF E1.L Q3 Up to 32 12
U.2 15mm NOW up to 32 12
AiC FHTQL up to 64 15
NGD Systems Computational Storage Solutions
June 11, 2019 NGD Systems, Inc. – Webinar 7
>10x faster Image
Classification
40% faster
1/3 the Space
Lower TCO
Native, Real-Time
Applications
500x Faster
Image Matching
OS running
On Drive
Presented by Thad Omura, EVP Marketing & Operations
scaleflux.com info@scaleflux.com /scaleflux
9Proprietary & Confidential
 Up to 8TB capacity
• Multi-NAND vendor
3D TLC
 GZIP, EC HW Engines
 Atomic Writes,
Namespace Streams
 Tunable latency &
performance
ScaleFlux Computational Storage
200%
Latency Consistency
Host FM & Tunable Performance
Customer Specific Workload
CSD vs. NVMe
200%
Queries Per Second
Atomic Write Support
Sysbench OLTP write-only
CSD vs. NVMe
161%
Jobs Completed
GZIP & EC Compute Engines
Teragen, Terasort
CSD+HDD vs. HDD only
3.0
*Applies to HDFS &
Up to 500%
GZIP Write Throughput
GZIP Compute Engines
FIO
CSD GZIP vs. CPU GZIP
260%
GZIP Write Throughput
GZIP Compute Engines
YCSB Load Benchmark
CSD vs. NVMe
Presented by Pankaj Mehra, VP - Product Planning Team
samsung.com /SamsungUS @SamsungUS
11 / 7C O L L A B O R A T E . I N N O V A T E . G R O W .
Samsung SmartSSD Technology Roadmap
Roadmap to smaller FF (U.2) and greater integration with SSD controller
Moves Data out ofSSDvia CPU &Memory
Limited scalability: Funnels 3 SSDs 1 FPGA
SmartSSD PoC
InternalDataTransfer
AddsBandwidthandProcessing
withoutFunneling
U.2 V1.0 SmartSSD
External FPGA
‘Q4’2018
HHHL AIC PoC
2H’2020 ~ 1H
2021
2nd Generation
2H’2019
U.2 MP
U.2FF:ScaleProcessingto24or48devices
2/4/8TB,25W,PCIeGen3x4,~300KLUTs
for accelerationkernels
NextGenSmartSSD
GreaterIntegrationand
MoreInternalBandwidth
12 / 7C O L L A B O R A T E . I N N O V A T E . G R O W .
Pushing Intelligence to Data
CPU
CPU
FPGA
/ GPU
PCIe
Today’s Architecture
Heavy domain-specific
compute in CPU
SSD
Large amount of
data transferred
SSD
SSD
TBs, PBs, EBs
dataset
PCIe
Storage and Data Acceleration
High, scalable
total internal BW
SSD
CTRL
FPGA
Accelerator
Offload,
UDF Accel.
&
Near-Data
Processing
Performance
# SmartSSD / serverNAND
Scale Performance with Data
 Concurrent processing on large amount of data
 Offloaded processing near the data
 HW acceleration of certain domain-specific fns.
 Immediate Insight from Data
Performance Ceiling
 Bottlenecks: Root complex, Ext FPGA
 Unnecessary data movement
 IO BW pressure on Host CPU, DRAM
SmartSSD
Performance
# SSDs / server
Today
CPU
Performance Ceiling
2x to 10x
DRAM
DRAM
$
CPU
CPU
DRAM
DRAM
$
13 / 7C O L L A B O R A T E . I N N O V A T E . G R O W .
TPC-DS Spark Q73 (Nested scan)
• Before:
• After:
Samsung SmartSSD Illustrative Performance Profile
Presented by Nick Adams, Board Member
snia.org /SNIA.ORG @SNIA
CONTACT US
Storageswiss.com
georgeacrump@storageswiss.com
@storageswiss
storageswiss
Will Your Backup
Architecture Meet
Tomorrow’s SLAs?
3 Steps to Make Sure!
For complete audio and Q&A please register for the
On Demand Version: bit.ly/ExtremePerform

Panel Discussion: Is Computational Storage a Better Path to Extreme Performance?

  • 1.
    Will Your Backup ArchitectureMeet Tomorrow’s SLAs? 3 Steps to Make Sure! For audio playback and Q&A go to: bit.ly/ExtremePerform
  • 2.
    George Crump, Lead Analyst& Founder Thad Omura, EVP Marketing & Operations Scott Shadley, VP of Marketing Pankaj Mehra, VP, Product Planning Team Nick Adams, Board Member
  • 3.
  • 4.
  • 5.
    what are someUSE CASES & SUCCESS STORIES?
  • 6.
    Presented by ScottShadley, VP of Marketing ngdsystems.com info@ngdsystems.com @NGDsystems
  • 7.
    Form Factor Available In 2019 Capacity (TB) MAXPower (W) M.2 22110 NOW up to 8 8 EDSFF E1.S Q3 up to 16 12 EDSFF E1.L Q3 Up to 32 12 U.2 15mm NOW up to 32 12 AiC FHTQL up to 64 15 NGD Systems Computational Storage Solutions June 11, 2019 NGD Systems, Inc. – Webinar 7 >10x faster Image Classification 40% faster 1/3 the Space Lower TCO Native, Real-Time Applications 500x Faster Image Matching OS running On Drive
  • 8.
    Presented by ThadOmura, EVP Marketing & Operations scaleflux.com info@scaleflux.com /scaleflux
  • 9.
    9Proprietary & Confidential Up to 8TB capacity • Multi-NAND vendor 3D TLC  GZIP, EC HW Engines  Atomic Writes, Namespace Streams  Tunable latency & performance ScaleFlux Computational Storage 200% Latency Consistency Host FM & Tunable Performance Customer Specific Workload CSD vs. NVMe 200% Queries Per Second Atomic Write Support Sysbench OLTP write-only CSD vs. NVMe 161% Jobs Completed GZIP & EC Compute Engines Teragen, Terasort CSD+HDD vs. HDD only 3.0 *Applies to HDFS & Up to 500% GZIP Write Throughput GZIP Compute Engines FIO CSD GZIP vs. CPU GZIP 260% GZIP Write Throughput GZIP Compute Engines YCSB Load Benchmark CSD vs. NVMe
  • 10.
    Presented by PankajMehra, VP - Product Planning Team samsung.com /SamsungUS @SamsungUS
  • 11.
    11 / 7CO L L A B O R A T E . I N N O V A T E . G R O W . Samsung SmartSSD Technology Roadmap Roadmap to smaller FF (U.2) and greater integration with SSD controller Moves Data out ofSSDvia CPU &Memory Limited scalability: Funnels 3 SSDs 1 FPGA SmartSSD PoC InternalDataTransfer AddsBandwidthandProcessing withoutFunneling U.2 V1.0 SmartSSD External FPGA ‘Q4’2018 HHHL AIC PoC 2H’2020 ~ 1H 2021 2nd Generation 2H’2019 U.2 MP U.2FF:ScaleProcessingto24or48devices 2/4/8TB,25W,PCIeGen3x4,~300KLUTs for accelerationkernels NextGenSmartSSD GreaterIntegrationand MoreInternalBandwidth
  • 12.
    12 / 7CO L L A B O R A T E . I N N O V A T E . G R O W . Pushing Intelligence to Data CPU CPU FPGA / GPU PCIe Today’s Architecture Heavy domain-specific compute in CPU SSD Large amount of data transferred SSD SSD TBs, PBs, EBs dataset PCIe Storage and Data Acceleration High, scalable total internal BW SSD CTRL FPGA Accelerator Offload, UDF Accel. & Near-Data Processing Performance # SmartSSD / serverNAND Scale Performance with Data  Concurrent processing on large amount of data  Offloaded processing near the data  HW acceleration of certain domain-specific fns.  Immediate Insight from Data Performance Ceiling  Bottlenecks: Root complex, Ext FPGA  Unnecessary data movement  IO BW pressure on Host CPU, DRAM SmartSSD Performance # SSDs / server Today CPU Performance Ceiling 2x to 10x DRAM DRAM $ CPU CPU DRAM DRAM $
  • 13.
    13 / 7CO L L A B O R A T E . I N N O V A T E . G R O W . TPC-DS Spark Q73 (Nested scan) • Before: • After: Samsung SmartSSD Illustrative Performance Profile
  • 14.
    Presented by NickAdams, Board Member snia.org /SNIA.ORG @SNIA
  • 15.
  • 16.
    Will Your Backup ArchitectureMeet Tomorrow’s SLAs? 3 Steps to Make Sure! For complete audio and Q&A please register for the On Demand Version: bit.ly/ExtremePerform