Chen Chen C

SSD Solution Architect, NSG, Intel Corp.

NVM Solutions Group

Legal Disclaimer

Intel may make changes to specifications and product descriptions at any time, without notice. Designers must not rely on the absence or characteristics of any features or instructions marked "reserved" or "undefined". Intel reserves these for future definition and shall have no responsibility whatsoever for conflicts or incompatibilities arising from future changes to them. The information here is subject to change without notice. Do not finalize a design with this information.

The products described in this document may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request.

Intel technologies’ features and benefits depend on system configuration and may require enabled hardware, software or service activation. Performance varies depending on system configuration. No computer system can be absolutely secure. Check with your system manufacturer or retailer or learn more at intel.com.

Intel disclaims all express and implied warranties, including without limitation, the implied warranties of merchantability, fitness for a particular purpose, and non-infringement, as well as any warranty arising from course of performance, course of dealing, or usage in trade.

Tests document performance of components on a particular test, in specific systems. Differences in hardware, software, or configuration will affect actual performance. Consult other sources of information to evaluate performance as you consider your purchase.

For more complete information about performance and benchmark results, visit http://www.intel.com/performance.

Intel does not control or audit the design or implementation of third party benchmark data or Web sites referenced in this document. Intel encourages all of its customers to visit the referenced Web sites or others where similar performance benchmark data are reported and confirm whether the referenced benchmark data are accurate and reflect performance of systems available for purchase.

Intel, Xeon, Xeon Phi and the Intel logo are trademarks of Intel Corporation in the U.S. and other countries.

*Other names and brands may be claimed as the property of others.

Copyright © 2016 Intel Corporation. All rights reserved.

2

NVM Solutions Group 3NVM SOLUTIONS GROUP

Cost Delay MORELESSHIGHERLOWER

Yesterday’s Storage and Memory Technologies Leave a Gap Performance or Capacity: Choose One

3

1.5 GBper day1

4,000 GBper day3

1,000,000 GBper day2

3,000 GBper day2

40,000 GBper day2

1) http://www.cisco.com/c/en/us/solutions/service-provider/vni-network-traffic-forecast/infographic.html2) http://www.cisco.com/c/en/us/solutions/collateral/service-provider/global-cloud-index-gci/Cloud_Index_White_Paper.html3) https://datafloq.com/read/self-driving-cars-create-2-petabytes-data-annually/172

3NVM SOLUTIONS GROUP

NVM Solutions Group 4NVM SOLUTIONS GROUP

Cost Delay MORELESSHIGHERLOWER

Intel Technology Innovations Fill the Memory and Storage Gap Performance and Capacity for Every Need

Intel® Optane™

Technology

Higherperformance

Intel 3D NAND

Technology

Lower cost & higher density

4NVM SOLUTIONS GROUP

NVM Solutions Group 5

Intel® Optane™ SSD with software

Redefining the memory / storage hierarchy

DRAM

NVDIMM

Intel® Optane™ ssdPcie*ssd

Pcie*ssd

SATA SSD

SATA SSD

HDD

HDD

tape

Massively Extend Memory

Persistent and Large Memory

Expanding data insights with:

Multiple affordable solutions

Intel Persistent memory 1

2

*Other names and brands may be claimed as the property of others.

NVM Solutions Group 6

Intel® Optane™ Technology is unique

Intel® Optane™ Technology is not NAND

bit addressable

Every memory cell can be individually addressed

revolutionary material

Most significant memory and storage advancement

in the last 20 years

write in place

Set or reset data as needed, no need to erase media

ultra-low latency

…together deliveringextremely fast media

NVM Solutions Group 7

3 reasons why the Intel® Optane™ SSd is the ideal caching solution

1. Responsiveness defined as average read latency measured at queue depth 1 during 4k random write workload. Measured using FIO 2.15. Common Configuration - Intel 2U Server System, OS CentOS 7.2, kernel 3.10.0-327.el7.x86_64, CPU 2 x Intel® Xeon® E5-2699 v4 @ 2.20GHz (22 cores), RAM 396GB DDR @ 2133MHz. Configuration – Intel® Optane™ SSD DC P4800X 375GB and Intel® SSD DC P3700 1600GB. Latency – Average read latency measured at QD1 during 4K Random Write operations using fio-2.15.

2. Source – Intel Data Sheet: Random/JEDEC up to 2.9 DWPD (5 Years) / 21.7 PBW, sequential workload up to 4 DWPD (5 Years) / 29.2 PBW3. Source – Intel: General proportions shown for illustrative purposes. Refer to slide 10 for a real-world example of cache to storage ratios

higher endurance+

Terabytes Written Specifications (TBW)2

Intel® Optane™SSD DC P4800X 30 D

WP

D

Intel® SSD DCP4600 (3D NAND) 2.9 D

WP

D

Intel® Optane™SSD DC P4800Xcache

more efficient=

Cache as a % of Storage Capacity3

Intel® SSD DCP4600 (3D NAND)cache

800

0

700

600

500

400

300

200

100R

an

do

m W

rite

MB

/s

Av

g R

ea

d L

ate

ncy

(us)

1000

800

600

400

200

0

Average Read Latency under Random Write Workload1

Intel® SSD DC P3700Intel® Optane™ SSD DC P4800X Random Write

lower and more consistent latency

8xlower

40xlower

Lower latency + higher endurance = greater SDS system efficiency

NVM Solutions Group 8NVM SOLUTIONS GROUP

Factors Affecting Data Center Storage Decisions

Reliability Performance Operating Costs Price

Failure RatesTransactionThroughput

Power

Data ProtectionReadPerformance

Cooling

Health Tracking Write IOPs Rack Space

TCO - Total Cost of Ownership

SSDHDD

NVM Solutions Group 9

Delivering an industry leading combination of low latency, high endurance, QoS and high throughput, the Intel® Optane™ SSD is the first solution to combine the attributes of memory and storage. This innovative solution is optimized to break through storage bottlenecks by providing a new data tier. It accelerates applications for fast caching and storage, increasing scale per server and reducing transaction cost. Data centers based on the latest Intel® Xeon® processors can now also deploy bigger and more affordable datasets to gain new insights

from larger memory pools.

World’s Most Responsive Data Center SSD1

1. Responsiveness defined as average read latency measured at queue depth 1 during 4k random write workload. Measured using FIO 2.15. Common configuration - Intel 2U PCSD Server (“Wildcat Pass”), OS CentOS 7.2, kernel 3.10.0-327.el7.x86_64, CPU 2 x Intel® Xeon® E5-2699 v4 @ 2.20GHz (22 cores), RAM 396GB DDR @ 2133MHz. Intel drives evaluated - Intel® Optane™ SSD DC P4800X 375GB, Intel® SSD DC P3700 1600GB, Intel® SSD DC P4600 1600GB. Samsung drives evaluated – Samsung® SSD PM1725a, Samsung® SSD PM1725, Samsung® PM963, Samsung® PM953. Micron drive evaluated – Micron® 9100 PCIe® NVMe™ SSD. Toshiba drives evaluated – Toshiba® ZD6300. Test – QD1 Random Read 4K latency, QD1 Random RW 4K 70% Read latency, QD1 Random Write 4K latency using fio-2.15.

NVM Solutions Group 10

Intel® Optane™ SSD Use Cases

DRAM

PCIe*

PCIe

Intel® 3D NAND SSDs

Intel® Optane™ SSD

Fast Storage and Cache

Intel® Xeon®

‘memory pool’DRAM

PCIe

Intel® 3D NAND SSDs

Intel® Optane™ SSD

DDRDDR

PCIe

Extend Memory

Intel® Xeon®

*Other names and brands names may be claimed as the property of others

NVM Solutions Group 11

DEMO - Optane SSD with IMDT

NVM Solutions Group

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

CN

IO node NVMeOF

IO node NVMeOF

IO node NVMeOF

IO node NVMeOF

IO node NVMeOF

Storage / Parallel

filesystem Lustre

HPC cluster Overview with a perspective of SSD

HighPerformance

tier

High Density Compute Nodeswith 2LM architecture

Flexible scratch space for small I/O

Improved bandwidth and I/O

responseGeneric Storage

NVM Solutions Group13

SSDs Help HPC to Boost Performance SSDs in Local Compute Nodes

Purpose: Scratch space to accelerate I/O intensive host workloads Checkpoint restart / Memory snapshot Decrease network utilization Improve KNL\FPGA performance Far memory tier for DRAM, Intel Memory Drive Technology Boot drive.

Technology Ingredients: 3D NAND PCIe* SSDs with NVMe* Intel® Optane® SSDs Low power products, M.2 PCIe and SATA

I/O Node, Burst Buffer and DTN Designs

Purpose: Accelerating data transfer in-and-out of the compute by

scaling I/O nodes in regards to compute nodes Improve small random I/O Cost saving to enable diskless compute nodes, maintaining

close to local I/O latencies

Improved HPC Storage Performance

Purpose: Bring NVMe into typical HPC storage ecosystem to improve

small I/O performance (HSM) Improving Lustre* performance for MDS and DSS features Improve cold storage CEPH and parallel FS such as BeeGFS

Technology Ingredients: Dual Port PCIe SSDs with NVMe designs for Lustre* MDS and OSS Single port SSDs with NVMe / SATA-based SSDs for scratch file

systems (no HA) Intel® CAS for DSS caching hints in Lustre*

Technology Ingredients: NVMeOF designs, NVMe-over-OPA designs All Flash NVMe burst buffers boost overall performance

*Other names and brands may be claimed as the property of others.See Appendix for footnotes.

NVM Solutions Group

768GB All DRAMIntel® Optane™ SSD + Intel® Memory Drive

Technology + 128GB DRAM

1 Source – Intel tested: Intel® Optane™ SSD + Intel Memory Drive Technology configuration – 2x Intel® Xeon® CPU E5-2699 v4 @ 2.20Ghz, Intel® Server Board S2600WT, 128GB DDR4 + 4x Intel® Optane™ SSD DC P4800X, CentOS 7.3.1611. All DRAM configuration – 2x Intel® Xeon® CPU E5-2699 v4 @ 2.20Ghz, Intel® Server Board S2600WT, 768GB DDR4 CentOS 7.3.1611. Test – Memory optimized GEMM (MKL), segment size 18689, factor 22, threads 42. Estimated results were obtained prior to implementation of recent software patches and firmware updates intended to address exploits referred to as "Spectre" and "Meltdown". Implementation of these updates may make these results inapplicable to your device or system. *Other names and brands names may be claimed as the property of others

Original GEMM application

Memory-Optimized application

14

GEMM* performance + Intel® Memory Drive Technology

2322GFLOPS 2605

GFLOPS2786GFLOPS

up to

1.1xup to

0.9xFaster matrix multiplication

with optimized data locality1Near-DRAM performance of

memory-optimized app1

vs. original GEMM app vs. memory-optimized app

DRAM-like Performance in Original and Optimized GEMM Applications

NVM Solutions Group 15

Scale of IMDT versus SWAP to 72 threads 50/50 read write

0

1000000

2000000

3000000

4000000

5000000

6000000

1 2 18 36 48 72

IMDT

Swap with P4800

Swap with P4500

NVM Solutions Group

Key Topic Areas

• Breakthrough Storage Technologies are unleashing vast performance potential in a range of forthcoming Intel® SSDs .

• NVM Express is the interface of choice for next generation storage technology

• Broaden your HPC Solution Portfolio with the Intel SSD Data Center Family of products.

Find out more at www.intel.com/ssd

Summary & Next Steps

16

Placeholder Footer Copy / BU Logo or Name Goes Here

NVM Solutions Group

Compute Node optimized architecture

Architecture Ingredients:

• Attach 1-2 local SSDs to the compute node directly

• Improve workloads characterized by large dataset or swap drive

requirements in Genomics, Fluid Dynamics, Computational chemistry,

Structural analysis, Finance, HPDA or generic app development.

• Implement Checkpoint restart/Memory snapshot for critical workloads

• Minimize I/O requirements to an external storage

• Opportunity for low power SATA/PCIe boot SSDs

Fabrics

Skylake - EPSkylake - EP

Block Diagram

Purpose

Accelerate HPC applications by providing close to memory local storage capabilities.

Use Cases

• Seismology Research Accelerated Using

Supermicro® SuperServers with Intel® NVMe

• Improving the accuracy of CGG’s seismic

modeling for upstream oil and gas

• Improving Genome Sequencing at German

Cancer Research Center

• Speeding up tracing applications at RWTH

Aachen University

TM

NVM Solutions Group

Improving Performance of HPC Storage

Architecture Ingredients:

Bringing PCIe*/NVMe* into typical HPC storage ecosystem to

improve small I/O

Intel PCIe SSDs in for Lustre*, BeeGFS*, CEPH*

Enabling PCIe/NVMe Dual Port drives’ topologies

Purpose

Accelerating traditional HPC parallel storage by introducing new features to improve small I/O Use cases

Lustre:

Metadata server (MDS)

HSM storage tier

DSS with CAS

All flash scratch or HA solutions

TM

SW

SW

PCIe SSDs with NVMe

Storage Controller

Storage Controller

*Other names and brands may be claimed as the property of others.

NVM Solutions Group

I/O node, Burst Buffer, NVMe over Fabric (NVMeOF)

Architecture Ingredients:

• PCIe IO rich nodes with balanced input-output configuration, so,

the internal IO processing capabilities can scale externally.

• Time to market to OmniPath and Ethernet products.

• Typical attach rate is 30 compute nodes to 1 burst buffer node

• Optimal NVMe SSD count depends on the used fabrics solution

and available bandwidth.

Fabrics

Skylake - EP

Fabrics

Skylake - EP

Block Diagram

Purpose

Accelerating data transfer in-and-out of the compute by scaling I/O nodes in regards to compute nodes with close to local I/O latencies

Use cases

• Burst Buffer implementation with Cray DataWarp nodes

• Data Transfer Node designs with Aspera* and Zettar* solutions

• NVMeF supports OPA fabric with 10.4 driver release and kernel 4.5+

TM

NVMe Host Software

Host Side Transport Abstraction

Controller Side Transport Abstraction

Fib

re C

ha

nn

el

Infi

nB

an

d*

Ro

CE

iWA

RP

Ne

xt G

en

F

ab

rics

NVMe SSD