1. technology and computing

The Machine:
The future of technology
patrick.demichel@hp.com
Hyperscale Division EMEA
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Agenda
1: Vision
2: The core technologies
3: The Machine
A new IT age is dawning: can you realize the benefits?
Security
Big Data
The Compute
opportunity
Cloud
Mobility
1972
.004 MIPS
1989
20 MIPs
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
2039
Millions of MIPs
services
2014
124,000 MIPS
Tsunami of data on the horizon
202X will be the decade of Extreme Data; massive compute is required for Extreme Analytics and simulation
Diverse
Users
BIG DATA
Ad Hoc
Questions
4
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
New Solutions
Are
Needed
Today’s computing infrastructure unable to keep up
/
You won’t be able to get more
capacity for less
By 2020
Internet of Things
Big Data will be too big to extract
meaning from
IoT “Things”
200
(1)
Billion
You won’t be able to move your data
from where it’s created – useful data
may get ignored or discarded
Connected
Devices
By the time you’ve analyzed your
data – it will be out of date
Smart
Meters
Your infrastructure will require
more resources than you can get
Securing your enterprise will take
more computing resources than you
have
Pervasive
Connectivity
Smart Device
Expansion
Explosion of
Information
30
(2)
Billion
1
(3)
Billion
… for 8
Billion (4)
(1) IDC “Worldwide Internet of Things (IoT) 2013-2020 forecast” October 2013. (2) IDC "The Digital Universe of Opportunities: Rich Data and the Increasing Value of the Internet of Things" April 2014
(3) Global Smart Meter Forecasts, 2012-2020. Smart Grid Insights (Zypryme), November 2013 (4) http://en.wikipedia.org
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
HP Labs Data-Centric Data Center : THE MACHINE
NO choice need for an holistic redesign for big impact
Novel system architectures and software
Return
Consume
Foundational technology breakthroughs
6
© Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Technologies for THE MACHINE
3 disruptive technologies to the rescue
Photonics
Special purpose cores
Massive memory pool
The Machine
8
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Photonic
Why photonics?
Photonics
100Gbps
c
be
e
or
s
ic
m
on
st
co
e
w
po
er
ic
on
tr
ec
el
iv
ct
fe
or
s
ay
ty
ef
si
en
rd
fo
w
al
1Gbps
1m
10
ni
ay
o
ot
cm
ot
ph
10Gpbs
ph
ni
s
ay
o
ot
w
al
ph
25Gbps
4m
10m
100m
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Katharine Schmidtke, Finisar
HP Photonics
Photonics
for faster data transmission and lower energy
30-fold more
bandwidth at one-tenth the energy
Transmit data using light for
1 -2 pJ/bit
All communications will be optical
12.5 Gbps
• Bandwidth scaling
• 30x improvement over copper
• Lower Power
• 10x improvement over copper links
• Improved airflow & cooling
• Equivalent cost
11
© Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
light
Photonics
Photonics destroys distance
12
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Ring Resonators
One basic structure, 3 applications
SiGe Doped
A modulator – move in and out of resonance to modulate light on
adjacent waveguide
A switch – transfers light between waveguides only when the
resonator is tuned
A wavelength specific detector - add a doped junction to perform the
receive function
13
© Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Microrings
Full link configuration
Unmodulated
light
Off chip
Laser source
(shared)
Advantages
Modulators wavelength specific, no
additional mux
Same ring structure used for drop
filters
Loss budget dominated by cost
Up to 64 wavelengths
Outstanding issues
Ring tuning
Thermal stability
14
Splitter loss
0.1 dB per
binary stage
Modulator array
5μm rings
Single mode fiber
10μm mode diameter
Delay 5ns/m
Loss 0.4dB/km
© Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Silicon ridge waveguide
0.5μm wide 4.5μm pitch
Delay 118ps/cm
Loss 0.3 - 1dB/cm
Fiber coupler if going off chip
Loss 1dB per connection
Detector & drop
filter array
HyperX1 networks
Fully connected sub-networks in multiple dimensions
High radix switch
Ring resonator
socket
– Superset of “flattened
butterfly” networks and
hypercubes
– Fully connected networks
offer lowest hop count but
limited scalability
– Multiple dimensions increase
scalability at the expense of
hop count
– Many alternate paths with
one or more additional hop
– Non-minimal routes required
for full bisection bandwidth
15 ©2010 HP Confidential
1.
“HyperX: Topology, Routing, and Packaging of Efficient Large-Scale Networks” Ahn et al., Supercomputing 2009
Conclusions:
• Integrated photonics has the potential to:
– Dramatically improve memory bandwidth
– Significantly improve many-core performance
– Reduce power
– Simplify programming
− All at the same time!
•Near term applications such as optical buses
− Add significant system flexibility
− Save latency and power
•Longer term give opportunity to rethink system arch
− New architectures & flexibility (e.g., optical buses)
– Disaggregation and dematerialization enablement
The memristor
Memory scaling challenges
DRAM
18
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Massive memory pool
Flash
Disruption #1: Non-volatile memories
Breakthrough in storage and memory technology
Store large amounts of data
permanently like hard disks, but
100,000 times faster, and at much
lower energy
19
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
UNIVERSAL MEMORY
Massive memory pool
A drastic reduction of the memory stack complexity and cost
But requires a complete software stack redesign to leverage the full potentiality of the new architecture
20
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Benefits of universal memory
Massive memory pool
Example: a database transaction
Traditional databases struggle with big &
fast data
90% of a database transaction is overhead
Memory-semantics nonvolatile memory:
up to 10x improvement
Btree
8.1%
Logging
21.0%
Locking
18.7%
Latching
10.2%
Buffer
manager
29.6%
Other
12.3%
Source: S. Harizopoulos, D. Abadi, S. Madden, and M. Stonebraker, “OLTP Through the Looking Glass, and What We Found There,” Proc. SIGMOD, 2008.
21
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Other NVM Technologies
Spin-Transfer Torque (STT-RAM)
22
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Phase-Change (PCM)
New Opportunities in 3D Integration
System-in-Package:
Combination of
3D die stacking and
2.5D interposer
integration
2.5D packaging on
Silicon Interposer
3D die stacking
23
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Nanostores: in-memory compute
Flat converged storage hierarchy with
compute colocation for
10x-100x improvement in
performance per Watt
Example: Matrix Computation
for (i = 0; i < n; i++)
sum += A[i][i];
Physical
CPU
Cache
Memory
Narrow Memory
Channel
Cache
line
24
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
The MACHINE
Application-focused silicon
Special purpose SoCs
System on a Chip (SoC)-based Server
Traditional Server Motherboard
ECC Memory
ECC Memory
ECC Memory
Processor
Processor
Processor
SoC
Mgmt
Interface
Custom
Accelerators
NIC(s)
Storage
Ctrlr
Production
Network
Storage
Southbridge
Storage
Ctrlr
Management
Logic
Mgmt
Network
26
Video
VGA
Console
NIC(s)
Production
Network
HDDs
Mgmt
•
Less general-purpose, more workload focused
•
Dramatic reduction in power, cost, and space
•
SoC vendors bring their own differentiated
features and opportunities to disrupt markets
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
HP Moonshot: the first step
*Source: HP internal research
27
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Special purpose SoCs
6 words to summarize the vision
Electrons
Compute
Photons
Communicate
Ions
Store
Not substitutional technologies
Holistic re-architecting to get all benefits
28
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
From nodes to racks to system to IoT
29
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
New architecture enables fundamental changes
Analytics solutions and visualization
Next-gen analytics algorithms
Distributed workloads and data management at cloud scale
Distributed data sources and systems
Relational Data
(Vertica)
Heterogeneous
cores
30
NoSQL
Files/objects MapReduce
(graphs, matrices,
(StoreAll)
(Hadoop)
documents, key-value…)
Pool of
non-volatile
memory
Stream & Log
Data
(ArcSight)
Nanostores
Converged System/Networking Fabric
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Unstructured data
(Autonomy IDOL)
Operating system +
orchestration +
continuous assurance
The Machine: towards a new computing paradigm
Value
Analytics and
visualization
Exabyte-scale
algorithms
Million-node
management
Machine OS
Data
Ultra-efficient
hardware
Security
built-in
from silicon
upwards
A Massive Mesh of intelligent objects
Use case: aircraft sensors
Internet-of-Things big data affects all industries
20 TB
2
20 terabytes of
information per
engine per hour
twin-engine
aircraft
3
three-hour average
flight duration
25,000
365
commercial flights
per day (USA)
days in a year
1,095,000,000 TB
1 ZB
33
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Use case: a mesh of connected aircrafts …
34
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Project
and
roadmap
Holistic, systematic & step-wise roadmap to revolutionary impact
Nanostores & compute hierarchies
in Data-Centric DataCenters
Project Moonshot: Gemini,
Discovery Lab, PathFinder
Converged infrastructure:
blades & modular datacenters
The Machine
HP Labs: blades++, power &
cooling, mchannels/mbrokers
35
HP Labs: blades, ensemble
mgmt, SoC aggregation, fabric
computing, new design models
HP Labs innovations for 10-100X disruptions &
new information-to-insight markets
© Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
Aspirational History
• SoC Partners selected for
co-development
• Machine OS development
begins
• Edge devices ship in volume
• Core Machines running real-world
workloads at scale
• Machine OS released
• Memristors begin sampling
• Physical infrastructure of Core
prototypes established
• Open Source Machine OS SDK and
emulators released
• ISV Partner collaborations begin
2014
2015
2016
2017
• Core devices at volume
• Machine available as
product, service, and
as a business process
transformation
2018
• Memristor DIMMs
launched
• Integrated core
technologies
demonstrated
• Edge devices begin sampling
• Machine OS enters public beta
36
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
2019
2020
Distributed mesh
compute goes
mainstream
The MACHINE …..
The Machine Webpage
The Machine 3 min video
Memristor Lab Tour
Photonics Lab Tour
HP Analytics Lab
HP Security and Cloud Lab
37
© Copyright 2014 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice.
This changes everything