Programmable Architectures for Communication Systems - PowerPoint PPT Presentation

About This Presentation
Title:

Programmable Architectures for Communication Systems

Description:

... computational entities that operate in a multi-threaded fashion. ... Co-design (Step 1) : recognise concurrent operations and optimise communication ... – PowerPoint PPT presentation

Number of Views:40
Avg rating:3.0/5.0
Slides: 26
Provided by: alant157
Category:

less

Transcript and Presenter's Notes

Title: Programmable Architectures for Communication Systems


1

Programmable Architectures for Communication
Systems
D. K. Arvind Institute for Computing Systems
Architecture, Division of Informatics, The
University of Edinburgh, Mayfield Road, Edinburgh
EH9 3JZ, Scotland. Email dka_at_dcs.ed.ac.uk
2
Edinburgh - The Capital City
3
Overview
  • University of Edinburgh
  • Division of Informatics
  • Edinburgh InfoLab
  • Research
  • Collaboration

4
University of Edinburgh
  • Founded in 1583
  • Student Population - 18,023
  • Undergraduate - 15,350 Postgraduate - 2,673
  • Staff - 6,649
  • Academic Staff - 3,312

5
Informatics is the study of the structure,
behaviour, and interactions of both natural and
artificial computational systems.(http//www.inf
ormatics.ed.ac.uk/)
Division of Informatics
  • Institute for Adaptive and Neural Computation
  • Centre for Intelligent Systems and their
    Applications
  • Institute for Communicating and Collaborative
    Systems
  • Institute for Perception, Action and Behaviour
  • Laboratory for Foundations of Computer Science
  • Institute for Computing Systems Architecture

6
Division of Informatics
  • Informatics_at_Edinburgh enjoys an international
    reputation for both its teaching and research
  • Only department in the UK awarded the top 5A
    research rating in Computer Science in Dec. 2001
  • UKs biggest department with 87 research-active
    staff and 165 PhD students
  • Edinburgh-Stanford strategic research partnership
  • Location of the National e-Science Centre
  • Awarded top Excellent rating in the Teaching
    Quality Assessment

7
The Future .
  • Proliferation of Peer-to-Peer computing
  • fundamental force of change and restructuring
  • Examples
  • Cybiko - P2P wireless networked games
  • Napster - P2P sharing of music
  • Freenet - P2P information store
  • DoCoMo P2P communication
  • Unregulated communications channels
  • ISM, UWB, free-space optics, ..
  • System-on-Chip components
  • banalisation of silicon technology
  • Silicon falling behind
  • storage bandwidth improving at a greater rate

8
P2P systems - Challenges
  • Portability - Java, .NET
  • Performance - exploit concurrency
  • Mobility - size and energy consumption
  • Flexibility - soft- and hard-programmability

9
Research Focus
  • To explore novel architectures for P2P systems
    using banalised technology, and enlighten future
    development of disruptive products and business
    change
  • Our research is seeking programmable solutions
    which
  • harness progress in (a) technology (b) theory
  • implement high-performance algorithms and
    applications efficiently

10
Disruptive technology opportunities
  • System Architectures to explore
  • Personal switch/P2P processor
  • Hubless, ephemeral, transient networks
  • Info-torch/Info-Klieg light
  • P2Pn libraryphonegaming

11
Trends in the silicon fabric
  • Convergence of transduction, communication and
    computation - heterogeneous systems with sensors
    and actuators
  • High performance computation at modest power
    consumption
  • Pre-designed IP blocks with different timing
    characteristics
  • The dominance of programmable fabrics - both
    soft- and hard-programmable
  • The complexity of the designs will demand novel
    architectures and design styles


12
The Die Area reachable in 1 clock cycle (1.2 GHz)
At 0.1um (1 Billion transistors) only 16 of the
chip is reachable in 1 clock cycle Dominance of
interconnect delays over computational
ones Network of Temporal Regions
13
Micronets - An alternative vision of Systems
Architecture
Micronet or Network-on-Chip a network of
entities which operate concurrently and
communicate asynchronously Fractal model of
system design network of sub-systems, down to
network of transistors Control is layered and
distributed locally - behaviour can be decomposed
to run on architectural clusters with the optimal
mix of computational elements A clean separation
between computation and communication, and,
behaviour and timing - leads to a compositional
design style
14
Behaviour-Architecture Co-design
  • Integration Platforms composed of networks
    (micronets) of heterogeneous computational
    entities that operate in a multi-threaded
    fashion.
  • Applications composed of software blocks some
    pre-defined, such as communication protocols
    others, more specific to the application.
  • Co-design (Step 1) recognise concurrent
    operations and optimise communication at
    different levels of granularity in the
    application and map them to the platform
  • Co-design (Step 2) explore the trade-off
    between programmability (both soft- and hard-),
    and performance (MOPS/mWatt) of the application
    running on the platform

15
The COMPASS Design Environment
  • Visualisation of energy and performance effects
    of compiler optimisations
  • Distributed simulation platform on a 16-node
    Beowulf cluster
  • Java or C applications
  • SSA intermediate representation
  • Soft- and Hard-programmability

16
Design framework for programmable multi-threaded
systems
17
A micronet-based multi-threaded architecture
18
Automatic Synthesis of Micronet Architecture from
Specification
  • void
  • Micronet(chan tinst Inst, chan tpc Pc, chan
    tregval RegDump, chan Word MemDump)
  • //Define channels
  • chan tinst ALUinst, MUinst
  • chan tpc ALUpc
  • chan tack ALUCUack, MUCUack
  • chan tregreq RegRequest
  • chan tregreturn Xout, Yout
  • chan tregval ALUXin, ALUYin, MUXin, MUYin
  • chan twriteback toReg, ALUWBout, MUWBout
  • chan bool KillBus
  • //Spawn linked Functional Units in Parallel
  • // clock 32
  • par
  • //Buffers for register requests
  • // clock 32
  • ControlUnit(Inst, Pc,
  • ALUinst, MUinst,

19
Power/Speed estimations on the M/T architecture
TPU 1
TPU 0
Overall
20
Power - Speed Tradeoff for Programs executing on
Micronet Architectures
21
Example of an Internet Appliance
  • Bluetooth-based system in VCC
  • Two physical objects the WAPmobile, and a WAP
    phone
  • The behaviour of an internet- and
    Bluetooth-enabled Basestation, and a
    Bluetooth-enabled robot is simulated in VCC
  • The WAP phone controls the robot in real-time
    via the VCC behavioural models

22
Proven Research Expertise in Systems Architecture
  • Programmable Architecture Design
  • Micronet-based asynchronous architectures
  • Java and C compilation for multi-threaded
    embedded systems
  • Applications include Bluetooth- and 802.11-based
    ones
  • Vertically-integrated environment (COMPASS) for
    energy-conscious, high-performance embedded
    system design
  • Industrial research partners
  • Well-endowed laboratory, including a 16-node
    Beowulf cluster for simulations and
    state-of-the-art EDA tools

23
Model for Collaboration
  • Feature set
  • Beyond the envelope research
  • Pre-competitive several industrial partners
  • industrial support funding, equipment, body
    swap,.
  • Successful Examples
  • Silicon Structures (Caltech 1977 - 81)
  • Berkeley Wireless Research Center (1998 - )
  • MIT Media Lab (1987 - )

24
Road Map
  • Creation of the Edinburgh InfoLab to research
    architectures for future P2P systems
  • 5 founding industrial partners/subscribers
  • 30 PhD students in the steady state
  • Partners contributions Two 4-year PhD
    studentships per year, cumulatively for 3 years
  • Interested? Email dka_at_dcs.ed.ac.uk
  • More details at http//www.dcs.ed.ac.uk/dka

25
Thank You
Write a Comment
User Comments (0)
About PowerShow.com