18
ADVANCING DATA CENTER AGILITY AT SCALE CHRIS BUERGER Head of RSD Customer Success Team DATA CENTER GROUP

ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

  • Upload
    others

  • View
    6

  • Download
    0

Embed Size (px)

Citation preview

Page 1: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

ADVANCING DATA CENTER AGILITY AT SCALE

CHRIS BUERGERHead of RSD Customer Success Team

DATA CENTER GROUP

Page 2: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

2

DATA IS DRIVING INNOVATION

AND BUSINESS TRANSFORMATION

MOVETOCLOUD

COMPUTING

50BCONNECTED

DEVICES

DELUGE OF

DATAGROWTH OF

AI AND

ANALYTICS

TRANSFORMATION

OF THE

NETWORK

Page 3: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

3

TOWARDS ZERO TOUCHAI IS BECOMING AN INTEGRAL PART OF DATACENTER COMPUTING

Page 4: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

4

2.28 3.02 3.965.24

7.04

12.86

1.1

1.5

2.13

4.38

0.91.07

1.32

1.64

2.03

3.17

0

5

10

15

20

25

2014 2015 2016 2017** 2018** 2020**

Inst

alle

d b

ase

in m

illio

ns

The Internet of Things (IoT)* units installed base by category from 2014 to 2020 (in billions)

Consumer Business: Cross industry Business: Vertical-specific

Insatiable User Demands1 Flat Infrastructure Budget2

CHALLENGE FOR TODAY’S DATACENTER OPERATORS

1. https://www.statista.com/statistics/370350/internet-of-things-installed-base-by-category/2. https://www.gartner.com/technology/research/it-spending-forecast/

CAGR ~40%

$0

$500

$1,000

$1,500

$2,000

$2,500

$3,000

$3,500

$4,000

$4,500

2016 YR 2017 YR 2018 YR 2019 YR 2020 YR 2021 YR 2022 YR

$ U

S B

illio

ns

Gartner’s forecast for 2018 worldwide dollar-valued IT spending growth increased 1.8% pts to 6.2%

Data Center Systems Software Devices IT Services Communications Services

+3.8% +6.2% +2.8% +2.8% +3.0% +3.0%

Page 5: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

5

THREE WAYS TO LOWER COSTS

Better Management

Lower CapEx & OpEx

with streamlined operations;

Saves time to focus on

customers & services

Resource Pooling

Maximize utilization of high-

value assets and improve agility

with dynamic composability

Storage Sled

Compute

Compute

Accelerator Sled

Physical

Pools of

Resources

Modular Refresh

Lower refresh costs by

independently scaling and

upgrading resources, while

always having the best tech

CPU & Memory

2.5 Years

Solid State Drives

4 Years

Accelerators

3.5 Years

SmartNICs

3 Years

Page 6: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

6

RESOURCE POOLING & DISAGGREGATION

Typical Hyperscale Intel® Rack Scale Design

Fixed ratio of

resources in

standard servers

Inefficient Scale Out

due to resource

overprovisioning

Less flexible in

expanding hardware for

new workloads and

resource profiles

Decrease

Costs

Increase

Agility

ComposableDisaggregated

Page 7: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

7

RESOURCE DISAGGREGATION

Storage Pooling over PCIe NVMe over Fabrics (Ethernet)

Two implementation options for hardware disaggregation

Lower

Latency

Greater

Scalability

Available NOW in Intel RSD 2.3Available NOW in Intel RSD 2.1

Fewer

ResourcesPotential storage resource

savings from disaggregated vs.

direct attach flash storage at the

same throughput level2

CUT YOURNVMeResources

40%Up to

1

NVMe Drives

1 Source: Flash Storage Disaggregation, Stanford University, Klimovic, Kozyrakis, et al, April 2016

Page 8: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

8

MODULAR REFRESH BENEFITS

Modular RefreshPotential refresh savings from

better component lifecycle

management made possible by

hardware disaggregation3

44%

Up

to

2

CUT YOUR Refresh CostsOut

of Cycle

Refresh

Savings

CPU & Memory

2.5 Years

Solid State Drives

4 Years

Accelerators

3.5 Years

SmartNICs

3 Years

Current Behavior¹ With Intel® RSD¹

31%

40%

29%

Sold or

Donated

Recycled or

Scrapped

Repurposed

Less than 30% of hardware is repurposed

1 Source: Intel conducted survey of server refresh behavior (n=235), March 120182 Source: Disaggregated Server Architecture Drives Data Center Efficiency and Innovation, Shesha Krishnapura, Intel Fellow and Intel IT CTO, 2017

Page 9: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

9

BETTER MANAGEMENT

Compute

Composed NodeAttached Resources

Compose hardware

resources “on the fly” for

specific workloads

Pooled System Management Engine (PSME)Firmware located in the Baseboard Management

Controller (BMC) of each hardware component

Accelerator Drawer

PS

MEStorage Drawer

PS

ME

Compute

PS

ME

Network

PS

ME

Intel® Rack Scale Design

Pod ManagerDynamically composes

resources into server

nodes from inventory

Orchestration Software

App 1 App 2 App 3

Intel® RSD software

functions include:

Resource DiscoveryAutomatically discover and store

hardware characteristics and

location for all your resources.

Node CompositionDynamically compose compute,

storage, and other resources to

meet workload specific demands.

Telemetry DataMonitor data center efficiency and

detect, diagnose, and help predict

resource failures.

Page 10: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

10

THE PATH TO INTEROPERABILITY

Proprietary Islands of Composable Systems will no longer be

Purchased

Prescriptive

Power of SDOEnablement

Tools & Models

Demonstrable

Proof

Plugtests

Page 11: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

11

INTEL® RSD WORKS WITH STANDARDS

Intel® Rack Scale DesignOpen Sourced Software

Physical ArchitecturesRack Dimensions, Board FF, Power Delivery

OCP ODCC19”

Industry Standards

Intel Pod Manager

Composed Node 2

Orchestration

App 1 App 2 App 3

Composed Node 1

Page 12: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

12

Better Management

Modular RefreshResource Pooling

THREE WAYS TO LOWER COSTS

Lower CapEx & OpExwith streamlined operations

Saves time to focus on customers & services

Maximize utilization of high-value assets and improve agility with

dynamic composability

Lower refresh costs by independently scaling and

upgrading resources, while always having the best tech

Source: Flash Storage Disaggregation, Stanford University, Klimovic, Kozyrakis, et al, April 2016

Source: SafeDX* Deploys Intelligent Data Center Management Solution to Save Data Center Power, SafeDX

/ Intel, February 2018.

44%Up to

CUT your Refresh

Costs

Out of Cycle Refresh Savings

Source: Disaggregated Server Architecture Drives Data Center Efficiency and Innovation, Shesha Krishnapura,

Intel IT CTO, 2017

NVMe Drives

CUT your NVMe resources

40%Up to

SAFEDXCoolingCosts1

17%

1. Source: SafeDX* Deploys Intelligent Data Center Management Solution to Save Data Center Power, Case Study published by Intel and SafeDX in February 2018

Page 13: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

13

CUSTOMER USE CASE

Novogene improved efficiency & flexibility while reducing their

overall TCO using Inspur’s InCloudRack with Intel RSD

Increased

Resource

Utilization

Decreased

Costs

https://www.intel.com/content/www/us/en/architecture-and-technology/rack-scale-design/novogene-use-case.html

Page 14: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

14

CUSTOMER USE CASE

Novogene improved efficiency & flexibility while reducing their

overall TCO using Inspur’s InCloudRack with Intel RSD

Increased

Resource

Utilization

Decreased

Costs

https://www.intel.com/content/www/us/en/architecture-and-technology/rack-scale-design/novogene-use-case.html

Page 15: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

15

INTEL® RSD-ENABLED SOFTWARE

Vendor* Status

Intel® RSD

Enabling: Enabled in OSP 13 release in Q2’18

rsd-lib up-streamed in Q4’17

enabled in official release in Q1’18 (Queens)

AWCloud solution enabled since Q1’17

Use case demo with SUSE CaaS Platform at

Kubecon in Dec ’17; white paper available

Canonical MAAS, Juju enabled since Q2’17

99Cloud solution enabled since Q2’17

AMI MegaRAC Composer enabled since Q4’16Orchestration

App 1 App 2 App 3

Intel® Rack Scale Design

Composed Node

+ +Compute

Disaggregated Hardware

Storage Drawer

Accelerator Drawer

Compute

Compute

Compute

*Other names and brands may be claimed as property of others. The Intel RSD ecosystem continues to evolve; for the most up-to-date status, visit www.intel.com/intelrsd

Page 16: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

16

INTEL® RACK SCALE DESIGN ROADMAP

2017 2018 Beyond2019

Compute Management

(Redfish)

Storage Management

(Swordfish)

Accelerator Management

(Redfish)

Network Management

(Yang-to-Redfish)

NVMe Storage

(PCIe)

NVMe Storage

(Ethernet)

FPGA Accelerator

(PCIe)

FPGA Accelerator

(Ethernet)

NIC GPGPU

OPEN MANAGEABILITY

RESOURCE POOLING

Page 17: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater

17

INTEL® RACK SCALE DESIGN

INTEL.COM/INTELRSD

LEARN MOREINSPUR InCloudRack

INSPURSYSTEMS.COM/PRODUCTS/INCLOUD-RACK

Page 18: ADVANCING DATA CENTER AGILITY AT SCALE€¦ · Storage Pooling over PCIe NVMe over Fabrics (Ethernet) Two implementation options for hardware disaggregation Lower Latency Greater