Next generation-cloud-networking-storage-peer-research-report
Cloud Storage: The Next 40 Years
-
Upload
it-brand-pulse -
Category
Technology
-
view
2.521 -
download
0
Transcript of Cloud Storage: The Next 40 Years
Cloud Storage: The Next 40 Years Frank Berry IT Brand Pulse
1
AI, OPaaS?
2
Network security alert! Changing
encryption algorithm.
3
I’ve developed a new method for compressing photos.
4
Cloud Storage Industry Road Map
Instrumented
Storage is instrumented and accessible anywhere.
Storage management tasks
performed automatically based on policies.
Self-Driving
Global storage, server, network & app service chains.
Artificial Intelligence allows storage
to recognize and respond to complex problems and
opportunities.
Bionic
Billions of data sources shared by governments and businesses.
Neural networks and deep
learning allows storage admin avatars to develop capabilities on
their own.
HYPERSCALES EVERYONE ELSE
5
2016 – 2026
Data
6
Gen Z
KPCB INTERNET TRENDS 2016
(Post-Millennials, iGeneration, Founders Generation, Plurals, Homeland Generation)
7
Image Apps Usage Continues to Rise
KPCB INTERNET TRENDS 2016
8
Video Views Growing
KPCB INTERNET TRENDS 2016
9
Video Evolution Accelerating
KPCB INTERNET TRENDS 2016
10
8K Ultra HD (9TB/hour)
11
3D ( 3 streams of 2K, 4K or 8K)
12
VR (n streams of 2k, 4k or 8k)
13
VR (n streams of 2k, 4k or 8k)
14
VR Shopping: eBay & Myer in Australia
15
VR Shopping: Singles Day, Nov. 11
16
VR Shopping: Select an Item
17
VR Shopping: Size the Item
18
VR Shopping: Try Different Colors
19
VR Shopping: Done? Watch a Movie
20
16,000,000 361,000,000
500,000,000
12,500,000,000
35,000,000,000
75,000,000,000
1995 2000 2005 2010 2015 2020
By 2020, 75 billion devices will be connected. In mature markets, people will have 5-10 things connected.
IoT
3B photos shared per day
21
22
IP Video cameras that double as access points
Temp Sensors
Shopper Handheld
Scanner
Employee Handheld Scanner
Asset Tracking
Sensor Data Analytics
Video Data Analytics
Video Management System
To App Servers in Data Center
Door Sensors
Video analytics provide
occupancy, dwell time and trip wire
by grid coordinates
23
Smart City of 1M = 200 Petabytes/Day
24
1M Self-Driving Cars = 4,000 Petabytes/Day
25
“In an autonomous car, we have to factor in cameras, radar, sonar, GPS and LIDAR – components as essential to this new way of driving as pistons, rings and engine blocks. Cameras will generate 20-60 MB/s, radar upwards of 10 kB/s, sonar 10-100 kB/s, GPS will run at 50 kB/s, and LIDAR will range between 10-70 MB/s. Run those numbers, and each autonomous vehicle will be generating approximately 4,000 GB – or 4 terabytes – of data a day. Every autonomous car will generate the data equivalent of almost 3,000 people. Extrapolate this further and think about how many cars are on the road. Let's estimate just 1 million autonomous cars worldwide – that means automated driving will be representative of the data of 3 billion people.” - Brian Krzanich , Intel
Data Doubling Every Two Years
26
5,000 Petabytes Per Year in 2020
2016 – 2026
Media (HDD & SSD)
27
1 Terabyte HDD vs. SSD
9.6TB SSD $3.83/GB
96TB HDD 67₵/GB
5.7x
28
163 156
164 168
160 166
177
120
147
157
139 136 136 133 140 142
138 138
147 141
125
111 118 115
100
0
20
40
60
80
100
120
140
160
180
200
Q1 2010 Q2 2010 Q3 2010 Q4 2010 Q1 2011 Q2 2011 Q3 2011 Q4 2011 Q1 2012 Q2 2012 Q3 2012 Q4 2012 Q1 2013 Q2 2013 Q3 2013 Q4 2013 Q1 2014 Q2 2014 Q3 2014 Q4 2014 Q1 2015 Q2 2015 Q3 2015 Q4 2015 Q1 2016
http://www.anandtech.com/show/10098/market-views-2015-hard-drive-shipments
All HDD Shipment Trajectory Millions
29
$-
$2,000
$4,000
$6,000
$8,000
$10,000
$12,000
$14,000
$16,000
$18,000
2012 2013 2014 2015 2016 2017 2018 2019
Enteprise HDD Revenue Enterprise SSD Revenue
Enterprise HDD & SSD Cross Over in 2017
30
2016
2026
HDD Road Map
31
SSD Road Map
60 Terabytes
2016
1 Petabyte
2026
32
NVMe (PCIe interface) SSD
33
NVMe Displacing SATA SSD
34
NVMe over Fabrics will Dominate SANs
Ethernet (RoCE)
Fibre Channel InfiniBand Next Gen Ethernet (iWARP)
35
Software Defined Storage (CapEx or OpEx) Best for Unstructured Data
Year 1 Year 2 Year 3 Year 4 Year 5
EMC VNXe 3200 $192,922 $213,468 $242,108 $308,338 $380,010
NetApp E2700 $60,206 $87,838 $144,718 $196,890 $261,622
NEC M110 $83,017 $103,285 $130,975 $178,055 $225,203
Seagate Ultra56 $55,979 $64,493 $87,555 $104,713 $123,774
SUSE Enterprise Storage $70,100 $76,579 $84,825 $96,238 $108,607
$-
$50,000
$100,000
$150,000
$200,000
$250,000
$300,000
$350,000
$400,000
Source: IT Brand Pulse
5 Year TCO: 250TB growing 25% per year
36
Pivotal time for HDD market. Flash market maturing fast.
HDD
Flash
The hard yards
Time to start thinking about the next big thing (Storage Class Memory)
Overnight success
Everyone Grows
2015
37
The Next Ten Years: Pro Flash PoV
38
60M/Year HDD
SSD
The Next Ten Years: Pro HDD PoV
39
60M/Year
HDD
180M/Year
2016 – 2026
Selling & Buying (CapEx & OpEx)
40
1998 2002 2006 2010 2014 2016
On Premise Storage (CapEx)
On-Premise Storage as a Service (OpEx)
Public Cloud Storage as a Service (OpEx)
Selling Storage
41
Buying Storage: Private STaaS The Next Best Thing in 2016
0% 10% 20% 30% 40% 50% 60% 70%
None of the above
Subscribe to off-premise public cloud storage, send our dataoff-site, and pay for what we use
Lease on-premise server, storage and networking capitalequipment and keep our data on-site
Subscribe to on-premise private cloud storage, keep our dataon-site, and pay for what we use
Buy on-premise server, storage and networking capitalequipment and keep our data on-site
If I had to choose one method of acquiring enterprise storage for my organization, I would choose to:
42
Buying Storage: Private STaaS The Next Best Thing in 2016
0% 10% 20% 30% 40% 50% 60%
Other
Public cloud block, file and object storage
Private cloud block, file and object storage
Traditional enterprise NAS and SAN
Which enterprise storage environment will offer best price/performance with many different storage workloads?
43
$-
$200
$400
$600
$800
$1,000
$1,200
$1,400
$1,600
$1,800
2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Equipment Not Financed
Equipment Financed
Equipment Finance Industry Size—Billions of Dollars
Financial Innovation a Key Dynamic for Storage in the Future
44
63% of IT financed in 2017
Innovative Financing Helps People Buy Cars
Car Prices Length of Lease Monthly Payments % Cars Leased
The average new car loan is 64 months and 27.5% of all new car purchases are leases, the highest level of leases since 2006
45
Driving Financial Innovation
46
On Premise-as-a-Service
$42,825 MSRP,7-speed automatic transmission, all season tires, daytime running lamps, rains sensor, glass sunroof, power seat and steering column with memory iPod media interface, Sirius satellite radio, Harmon Kardon sound system, heated front seats, 17” spoke wheels, AMG sport line package
$1.50
Pay only for the miles you drive
Per Mile
$0 DUE AT SIGNING
47
On Premise Storage-as-a-Service
Off Premise Storage-as-a-Service
Game Changer
48
10 Year Data Center Storage Revenue Forecast
49
$-
$10,000
$20,000
$30,000
$40,000
$50,000
$60,000
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024
Traditional Enterprise Storage Capex
Public Cloud Storage OpEx
Private Cloud Storage OpEx
84%
4%
12%
50%
23%
27%
On premise and off premise
Billions
Financial Innovation Needed to Bridge Vendor Beauracracy and Customer Demand
50
Oracle
Credit
NETAPP FINANCIAL
Hewlett Packard
FINANCE
Financial Services
Financial Services Finansal Hizmetler
1998 2002 2006 2010 2014 2016
On Premise Storage (CapEx)
On-Premise Storage as a Service (OpEx)
Public Cloud Storage as a Service (OpEx)
Landscape When OPaaS Takes Off
51
2016 – 2026
Storage Brands
52
Storage Branding 2016
Ingredients On Premise Brand
53
Storage Branding 2026
Ingredients Cloud Brand
54
Gartner Magic Quadrant: Cloud Storage
Belongs here
2014 AT&T #3
2015 HP Out
2016 Oracle In
55
2016 Cloud & On-Premise Voted by IT Pros
Cloud Backup & Archive as a Service AWS AWS AWS AWS AWS AWS
Cloud Dedicated Enterprise Servers AWS AWS AWS AWS AWS AWS
Cloud Enterprise Class Storage AWS AWS AWS AWS AWS AWS
Cloud SSD Storage AWS AWS AWS AWS AWS AWS
Cloud Virtual Desktops VMware VMware VMware VMware VMware VMware
Cloud Virtual Servers AWS AWS AWS AWS AWS AWS
Cloud Unified Communications Microsoft Microsoft Microsoft Microsoft Microsoft Microsoft
On-Premise vs. Cloud Backup & Archive EMC & Veeam (tie) AWS EMC EMC AWS AWS
On-Premise vs. Cloud Dedicated Enterprise Servers
HPE AWS HPE HPE AWS HPE
On-Premise vs. Cloud Enterprise Class Storage
EMC AWS EMC EMC AWS AWS
On-Premise vs. Cloud SSD Storage EMC Pure Storage EMC EMC EMC EMC
On-Premise vs. Cloud Virtual Desktops VMware VMware VMware VMware VMware VMware
On-Premise vs. Cloud Virtual Servers VMware VMware VMware VMware VMware VMware
On Premise vs. Cloud Unified Communications
Cisco Cisco Cisco Cisco Cisco Cisco
Zadara Storage
57
eCommerce Consolidation on IaaS
58
eCommerce Brands Will Look Like This
59
Mobile Clients DC Infrastructure
DC IaaS
Carriers
Media
SaaS
Tectonic Shifts in IT Brands
Self-Driving
eCommerce (Retail, Healthcare, Banking)
X
60
2016 – 2026
Storage Technology (Artificial Intelligence)
61
From Mad Scientists to IT for the Masses
NASA Spin-Off Technologies Infrared ear thermometers
Artificial limbs Solar Cells
Freeze drying Scratch-resistant lenses
Water purification Portable cordless vacuums
Powdered lubricants Space blanket
Aircraft anti-icing systems Video enhancing and analysis systems
Firefighting equipment OpenStack
Software catalog
Hyperscale Spin-off Technologies Open APIs top to bottom
OCP Servers OCP Storage
OCP Switches Software Defined Data Center Software Defined Networking
Software Defined Storage Scale-out Storage
Scale-out Computing Object Storage
NoSQL Databases Hadoop
Containers Artificial Intelligence
62
Machine Learning as a Service
63
Augmented Humanity: Integration of digital technologies with human biosystems
64
Direct Neural Interfaces
Artificial Intelligence, Machine Learning & Deep Learning
65
Perform tasks that normally require
human intelligence, such as visual
perception, speech recognition,
decision-making, and translation
between languages.
Take neural networks, make them huge, increase the layers and the neurons, and then run massive amounts of data through the system to train it.
Rather than hand-code software routines with a specific set of instructions, use algorithms to parse data, learn from it, and then make a determination or prediction about something in the world.
Noun A computer system modeled on the human brain and nervous system. Simulates the behavior of biological neural networks, as in pattern recognition, language processing, and problem solving, with the goal of self-directed information processing.
neu·ral net·work
66
Inspired by The Brain
What separates the future of computing from the past
67
Neuromorphic Chips
68
Applications for Deep Learning
Computer Vision
Natural Language Processing
Speech/Audio Processing
Touch
Smell
Taste
69
70
Electronic Nose
71
72
73
Conversation
74
Vision
75
Save Energy
76
Diagnose Cancer
77
“Can we do better for human well-being if information is more broadly accessible, where you’re leveraging the mindshare of the entire planet to evolve the models of disease?” Schadt asks. “Absolutely.” This is medicine as math, not guesswork, and every disease—even stage 4 cancer—might one day be druggable. - Wired
The Cure for Cancer Is Data—Mountains of Data
78
Voice User & Admin Interfaces
KPCB INTERNET TRENDS 2016
Network security alert! Changing
encryption algorithm.
KPCB INTERNET TRENDS 2016
79
Voice Gaining Search Share
KPCB INTERNET TRENDS 2016 80
Person to Machine Voice Interaction Adoption Keys
As speech recognition accuracy goes from say 95% to 99%, all of us in the room will go from barely using it today to using it all the time. Most people underestimate the difference between 95% and 99% accuracy – 99% is a game changer... No one wants to wait 10 seconds for a response. Accuracy, followed by latency, are the two key metrics for a production speech system... ANDREW NG, CHIEF SCIENTIST AT BAIDU
81
Machine Speech Recognition @ Human Level for Voice Search in Low Noise Environment
KPCB INTERNET TRENDS 2016
82
What I do remember is that Cisneros told the joke in Spanish, and I heard it in English. The demo’s still early: it only worked in one of my ears, and the joke’s punchline didn’t come through for a really awkward five seconds or so. But it felt like the beginning of something big. - Wired
83
Self Driving Cars
KPCB INTERNET TRENDS 2016
84
A Metric for Success: Miles Driven
KPCB INTERNET TRENDS 2016
85
Self-Driving Storage D
escr
ipti
on
Ex
amp
le
Tim
e Fr
ame
L0 L1 L2 L3 L4
No Automation Function-Specific Automation
Combined Function Automation
Limited Self-Driving Automation
Full Self-Driving Automation
Admin in complete and sole control of primary storage controls (media, volume management, backup, replication, dedup, compression) at all times.
Automation of one or more primary storage control functions, but no combination of systems working in unison
Automation of at least two primary storage control systems working in unison
Admin able to cede full control of all performance and HA critical functions under certain conditions. Admin is expected to be available for occasional control, but with sufficiently comfortable transition time
Self-driving storage can perform all performance and HA-critical driving and monitoring functions 24x7, 365 days a year
Since enterprise-class storage invented - 1960s
Virtualization (RAID) Automated backup Storage monitoring
HSM Storage migration QoS
AWS, Google, Azure Service Chains Artificial Intelligence
ITBP – Self Driving Storage System Classifications
1990s - Today 2000s - Today 2010s - Today 2020s
86
A Metric for Success: Admins Per Server
20,000 Servers per Admin
Hyperscale Enterprise
500 Servers per Admin
Servers are fully instrumented and integrated with analytics and automated
server management software
Best-in-class VMware environments
87
“In playing StarCraft, the AI system will have to come up with real-time strategies for choosing one of the three distinct races at the beginning of the game; choosing when and how to farm minerals and gas; deciding when and which buildings and units to construct; and scouting unseen areas of the map and remembering that navigational information over the course of the game. An AI engine would therefore have to make use of the skills of memory, mapping, long-term planning, and adapting to changes in plans using information that is continually being gathered, which translates to hierarchical planning and reinforcement learning.”
Blizzard and DeepMind have created an open test environment within the StarCraft II game for artificial intelligence researchers to use worldwide.
88
“In deploying enterprise STaaS, the AI system will have to come up with real-time strategies for choosing one of the three distinct platforms (public, on-premises or hybrid) at the beginning; choosing where, when and how to implement service levels; deciding where, when and which storage media to deploy; and scouting unseen areas of the application environment and remembering that navigational information over the course of the service agreement. The Zadara AI engine would therefore have to make use of the skills of memory, mapping, long-term planning, and adapting to changes in plans using information that is continually being gathered, which translates to hierarchical planning and reinforcement learning.”
Zadara Enterprise Storage-as-a-Service 2026
89
Goal-Based AI
VPSA
VPSA
VPSA
90
91
2026 - 2056
92
Electricity Power Medicine
Pulleys Levers
Engines
Lighting Telephone Television
Computing
Artificial Intelligence
93
7 million BC (last common ancestor)
1960 1980 2016 2056
Evolution of Storage Technology
2500 BC
Written Mainframe Client-Server Cloud Bionic
94
Anatomy of a Software Defined Data Center
Facility Circulatory system
Servers Nervous system
95
Dendrites pick up signals from other Neurons
Axons (white matter) are long nerve fiber that conduct information from the cell body in the form of an electrical impulse
The Cell Body is the neuron’s powerhouse, responsible for generating energy and synthesizing proteins
Axonal terminals are end points of an axon’s branches, where electrical impulses are discharged; releasing neurotransmitters to other cells’ dendrites
100 Billion Neurons (Servers) in Each of Our Brains
Source: National Geographic, The New Science of the Brain
10,000 synapses per neuron 100 trillion to 1 quadrillion connections
96
Densely Packed (in the Data Center)
100 Microns 10 Microns 3 Microns
Dendrites Cell Bodies Dendrites Axons Dendrites Axons
Source: National Geographic, The New Science of the Brain 97
Massive Parallel Processing Before Output
Many receptors Many transmitters
98
A helmet of sensors at the Martinos Center for Biomedical Imaging – part of a brain scanner requiring almost as much power as a nuclear submarine. Antennas pick up signals which computers convert into brain maps (see next slide).
Source: National Geographic, The New Science of the Brain 99
The brain’s many regions are connected by 100,000 miles of fibers called white matter—enough to circle the Earth four times. Images like this reveal the specific pathways underlying cognitive functions. The pink and orange bundles transmit signals critical for language.
100,000 Miles of Network Cable
Source: National Geographic, The New Science of the Brain 100
1 Billion Terabytes of Storage
450,000 terabytes (.45 exabyte)
Storage capacity needed to produce mouse brain image
1.3 billion terabytes (1,300 exabytes)
Storage capacity needed to produce human brain image
Source: National Geographic, The New Science of the Brain
Almost 3x data to be produced in 2020
101
20 Watts of Power
Ceiling fan 24 watts
CFL light bulb 18 watts
Desktop & Monitor (in sleep mode)
1-20 watts
Nintendo Wii 18 watts
Human Brain 20 watts
102
Server Configured Like a Human Brain
1.3 Billion Terabytes of Storage
10,000 Network Ports 100,000 Miles of Cabling Included
>100 PetaFLOPS Processing Power
20W Power
77 Cubic Inches 19” Wide x 1.75” (1U) High x 2.3” Deep
3.3 Pounds
103
World’s Most Powerful Supercomputer Sunway TaihuLight
93 Petaflops 40,000 Nodes
10M Cores 1.3PB Memory 15MW Power
2,250ft2 Floor Space
104
>100 Petaflops Needed to Simulate 1 Mouse Brain
Sunway TaihuLight World’s Most Powerful
Supercomputer
GigaFlops
TeraFlOPS (1012)
PetaFlOPS (1015)
ExaFlOPS (1018) ZettaFlOPS (1021)
YottaFlOPS (1024)
105
SW and HW vision, hearing, language, etc., receptors
High bandwidth, low-latency networks
Scale-out servers
SW and HW vision, hearing, language, etc., transmitters
We’ve just started
Source: National Geographic, The New Science of the Brain 106
SW Instrumentation: Open APIs Up in Cloud Platforms
Nova Neutron Cinder Swift
RESTful
Horizon
AWS
107
Co
mm
on
Op
era
tio
ns
Business-specific Applications
Business-specific Services, Web Frontends
Connectivity
Clo
ud
Fo
un
dat
ion
Co
mm
on
Se
curi
ty
Business-specific Applications
Field Devices - Gateways, building
servers, bridges, ‘modules’
Apps - mobile apps for control,
commissioning, offline (non-cloud) connected
Services - Cloud-based software
services, application back-
ends
Apps - Mobile apps, web apps, desktop
apps for cloud-connected
experiences
Applications
Data & Analytics
Connectivity
Business
Support
Common UI Applications
Business-specific Data, Algorithms, Workflows, Dashboards
Data & Analytics IoT Device Management
Domain & Application Services
Pla
tfo
rm S
cop
e
Bu
sin
ess
Sco
pe
SW Instrumentation: Open APIs Down to Light Bulbs
108
HW Instrumentation: Networking
FPGAs generate and consume their own networking packets independent of the hosts, Every FPGA in the datacenter can reach every other one (at a scale of hundreds of thousands) in a small number of microseconds, without any intervening software. FPGA resources are an independent computer in the datacenter, at the same scale as the servers, that physically shares the network wires with software.
Appliance
Server CPU
FPGA (NIC)
Microsoft Project Catapult
109
HW Instrumentation: Graphics Processing
110
Brain-Computer Interface (BCI)
111
“Move arm” “Move cursor”
112
Bionic Brain
“People with spinal cord injuries can’t move because the brain and body no longer communicate. Scientists hope to restore motion with a mechanical skeleton controlled by the wearer’s thoughts. It’s a daunting challenge: Hundreds of sensors must be implanted in the brain to send commands to the exoskeleton. Signals must also travel in reverse, from touch sensors telling the brain where the body is in space.”
113
Huawei Enterprise Confidential Page 114
Avatar Project Milestones
Avatar D 2040- 2045 A hologram-like avatar
Avatar C 2030- 2035 An avatar with an artificial brain in which a human personality is transferred at the end of one’s life
Avatar B 2020- 2025 An avatar in which a human brain in transplanted at the end of one’s life
Avatar A 2010- 2020 A robotic copy of a human body remotely controlled via BCI
Brain to Brain Interface (B2B)
115
116
117
Can we do that?
Yes
No Absolutely
118
$$
Thank-You [email protected]
119
Deep Learning Basics
120