Stephen Worn CTO, Data Center Dynamics DATA CENTER TRANSFORMATION in the Cloud Era • It’s ALL about the App • The Cloud Paradigm • Next Generation Data Centers • Hyperscale • Enterprise • Multi-Tenant Data Centers (MTDC) • Modular ITC • System Interconnects and HyperConvergence • CAPEX and OPEX Opportunities • Closing and Questions Holistic Approach

NETWORKING INFRASTRUCTURE

APPLICATIONS A DATA CENTER’S MISSION SHOULD BE TO CREATE RELIABILITY, MITIGATE RISK, AND PROVIDE UPTIME FOR THE TECHNOLOGY AND

APPLICATIONS THAT IT ENABLES

ENVIRONMENT

CAPEX/OPEX 70%

2012 2015 60%

60%

Virtualisations

50% Upgrades

43%

Cloud 37% 40%

Service type

34%

32% environment

27%

30% 26%

25%

% % Sample

23% 23%

22% 22%

21%

20%

19% 18%

20% 18%

15% 15%

14%

13%

11%

10%

8% 8%

10% 6% 5%

0%

IAAS

SAAS

PAAS

Users

Public

Server

Hybrid

Servers

Private

System

Storage

Network utilities The technologies that enable this are based on networks linking data centers and linkingApplications

active IT equipment within a data center Software defined Software These demands mean the need to invest to improve networking performance has increased from 2013 to 2016 2013 2014 2015 2016 Increased IT capacity requirements 35.4% 39.6% 42.2% 48.6% To reduce operating costs 39.8% 38.6% 39.5% 45.3%

To enable virtualization / cloud 25.6% 29.6% 32.5% 42.3% computing, service development

To be 'greener' & more sustainable 29.8% 30.5% 32.5% 36.2% To improve network performance 21.3% 26.5% 31.2% 35.4% To improve space use 23.3% 25.4% 28.4% 35.1% To improve security 24.5% 28.6% 32.5% 35.1% To increase power into facility 32.5% 30.8% 31.1% 35.1% To increase redundancy 29.6% 28.6% 31.5% 33.4% An average 1 in 3 nomination = once every 3 years = frequency of IT refresh What are the business drivers underpinning all of this? Challenges Force Change • Increasing power densities (typically 4-6KW, now 12 up to 30, with possibilities of 70kW per rack) • Improved cooling strategies • Need for scalability • Need for modularity • Data center consolidation • Server virtualization and demand from DevOps • Energy saving initiatives and regulations

Apps Moving to the Cloud

Business Apps

Consumer Apps Business and Life in Cloud

23 billions smart devices by 2020 HPC maximum flops ×2 annually 15 Internet of things 93 petaflops/s (10 ) in 2016 Sunway TaihuLight (China)

Every 200 mobile devices = Social network: 1 more server in the cloud 300 million pictures/day uploaded to Facebook

Source: 1. Cisco Visual Networking Index Forecast 2. Garner Worldwide IT Spending 3. Top 500 list Top 10 change 4. Gizmodo/Facebook Next Generation Cloud Ecosystem Traffic Moving to the Cloud

Traffic Growth Traffic Destination

Data Center to User 18% 25% CAGR 83% Data Center 7% 61% to 39% 17% 5% CAGR Data Center 75% Within Data 2014 2019 Center Traditional Cloud

Dell’Oro, 2015 Dell’Oro, 2015 The New Cloud Paradigm

Belden 2016 Hybrid Cloud Solutions

Belden and Anixter 2016 Hyperscale Data Centers

Market Size is estimated to grow from $71.2 billion by 2022, with a CAGR of 20.7% from 2016 to 2022 Techresearchandanalysis, 2016 Source: Microsoft (Dublin, Ireland)

Source: Google Source: Apple New Cloud Challenges – Move to Edge Computing

Belden and Anixter 2016 Is Scale the Answer to Everything?

• Diminishing ROI for further scale increases… – Cost, latency and power consumption

• Resilience requirements – ITC DCs closer to the end users, and duplication

• Evolving workloads – More flexible orchestration and edge computing

Belden and Anixter 2016 Understanding ‘As A Service’ Data Centers

Enterprise Multi-Tenant Hyperscale Owned

Infrastructure Internally Managed IaaS as a Service Traffic Software Software

SaaS as a Service SaaS as a Service Platform FaaS DCaaS PaaS as a Service

Assets Assets Content Delivery Network Services

DCD Intelligence CDN Report 2015 and Belden 2016 Over-The-Top Data Traffic Global Internet traffic Upstream Global Internet traffic Downstream

BitTorrent Netflix YoutTube YouTube 18% Amazon Video 26% Netflix 25% SSL - Other 35% HTTP - OTHER Google Cloud iTunes 2% 13% iCloud 2% Hulu 3% HTTP - Other 2% SSL - Other 3% Face book 2% Xbox One Games 10% Facetime 2% Facebook 4% 6% 3% 18% Skype BitTorrent 7% 9% 3% 4% The Others 4% The Others

Top 10 Peak Period Applications, North America, Fixed Access Source: Sandvine 2016 Over-The-Top Data Traffic Global Internet traffic Upstream Global Internet traffic Downstream Facebook YouTube 15% SSL - Other Facebook 21% Google Cloud 25% HTTP - OTHER 34% HTTP - Other SSL - Other 14% YoutTube Instagram Snapchat 2% Snapchat Instagram 14% Netflix 3% BitTorrent iTunes 9% 3% 2% FaceTime Google Cloud 4% 9% 2% 9% iCloud 7% MPEG - Other 5% 5% 7% 2% 3% The Others The Others 4%

Top 10 Peak Period Applications, North America, Mobile Access Source: Sandvine 2016 Latency Impacts

Source: EdgeConnex @ OFC 2016 Multi-Tenant Data Center Service Types

Belden and Anixter 2016 New Data Centers Ecosystem at the Edge

• Colocation/Edge data centers • The emergence of secondary data − Caching content closer to the end user center markets MTDCs, CDNs − Smaller and more distributed locations • Provide direct connects – Being part of the Cloud value chain − Enterprise end users prefer certain mission critical elements of their IT infrastructure to – Direct connectivity to Enterprise be physically closer to their offices – Consumption-based pricing models − Edge-Computing nodes allow extending the – No more long term commitments – Cloud and support low latency applications month to month pricing

Enterprise Data Center – Still a Significant Market

• Enterprise Global data center space in square feet – Financial institutes Enterprise MTDC Cloud – Health care 8% 10% – Storage, IT infrastructure 17% 19% – Small-medium enterprise

• Government, research/education 76% 71% – High performance computing – Big data analytics – Internet of things Source: 451 research 2015 2017 Server Shipment Forecast Server shipment split • Current installed base 100% – Enterprise >> Cloud 80% • Migration trend 60% – From Enterprise to Cloud service provider 40% – From Enterprise to 20% MTDCs for Hybrid 0% Cloud service 2015 2016 2017 2018 2019

Enterprise Cloud SP Source: Dell’Oro 2015 Traditional DC VS Modular DC

Traditional Datacenter Modular Datacenter

• Complex engineeringlong term construction, • Standardized Design and Engineering average 2 years • Shortened Construction Period – 1 to 3 months • Low infrastructure utilization, stranded • High infrastructure utilization capacity and PUE>2.0 • Balanced PUE<1.6 Based on IT Capacity • One-time construction, large Initial • Energy Efficienct: Improved power consumption investment, Highly managed ROI and modular subsystems. • • Unused capacity high Opex and Stranded IT Concentrated power & cooling improve infrastructure utilization • Fixed infrastructure structure, scalable，can • Flexible structure, phased deployment be slow to business reacting • Financial benefits, lower CAPEX and OPEX across all environments and subsystems • Subsystem cross-management weaknesses across multiple systems. high cost of O&M • Capacity and Location On Demand • Decentralized Control with Centralized Mgt.

Schneider Electric and Rittal 2016 • DCIM and Predictive Analysis Modular data center has been widely accepted

Google Oregon DC HP launched 40 ’POD

Microsoft Chicago DC

SUN Black box IBM launched PMDC SGI Ice Cube Tencent modular DC Cisco launched UCS

Year 2005 2006 2007 2008 2009 2010 2011 2012 2016 Modular Spaces • Modular design is slightly different thinking – It can be as simple as a cabinet that contains individual modules or bays – Or as dynamic as an entire data center space • Modularity could mean the ability to expand in a planned or standard sets – Converged Solutions – kW load – Compute/network/storage • Modular spaces need to provide: – Support future technologies – Support cooling and power requirements – Increase energy efficiency – Create potential “Cookie Cutter” design, but matched to the building and customer needs

DCD Intelligence Modular Report 2015 Modular Design Challenges

• Will it be flexible? – Designs are usually completed 3-4 years prior to product install • Can it accept future technologies? – Equipment is changing from power requirements to cooling requirements – Ex. New Switches and Air path – Easily upgradeable • Network Connectivity – Can it support current and future network demand and connections • Space – Newer devices requiring greater space footprint, power and cooling.

DCD Intelligence Modular Report 2015

HyperConvergence 8 MONTHS in development 18 MONTHS in development 43 MONTHS in development Legacy Stack Convergence 1.0 Convergence 2.0 Convergence 3.0 Integrated Systems Partial Convergence Hyperconvergence

Servers + VMware

Storage Switch Pre-integrated storage and server Converge only HA Shared Storage storage and server Hyperconverge SSD Array entire stack Backup Appliance WAN Optimization Cloud Gateway Storage Caching Best of Breed Backup Apps Cloud Enterprise Economics Capabilities 1. Data Protection 2. Data Efficiency 1. Low Cost 3. Performance 2. Commodity x86 4. Global Unified 3. Scale-out Management Source: Simplivity 2016

Application/Cloud High Speed Connectivity Challenges

2016 BSRIA, System Interconnects Speed & Media BSRIA, 2015 Network Speed Technology Lifecycle

Operating bandwidth Speed 1G 10G 25G 50G Distance100G limitation 200G 400G 1T Power consumption

Twisted- Pair Distance limitation

DAC Parallel optics Distance limitation (Twin-ax) Infrastructure cost SWDM

MMF Active equipment cost Silicon Photonics

SMF

Source: BSRIA, Belden, Anixter Enterprise Switching Architecture ToR Switch to Server Connections 5% EoR 33% 42% Twisted MoR Topology Reach DAC AOC MMF SMF ToR Pair Centralized 5% 15% EoR ToR 3 m   Other MoR Adjacent 5 m Rack    6% Centralized 18% MoR / EoR 30 m    Other 7% 66% Centralized 100 m    3%

MTDC Switching Architecture BSRIA, 2015 Data Center Migration Paths BSRIA, Belden, 2015 and 2016

BASE-SR4 BASE-SR4 BASE-T BASE-T/CR BASE-SR4 BASE-CR BiDi PSM4 (DR4) /PSM4 (DR4) /CWDM4 10G 40G 10G / 25G 40G / 100G 25G / 50G 100G / 200G / 400G

1G 10G 1G 10G / 40G 10G 40G

Access Aggregation Access Aggregation Access Aggregation Enterprise Hyperscale MTDC DC DC Closing and Questions. Thank you