aiming at the natural equilibrium of planet earth requires to reinvent computing reiner hartenstein...

Aiming at the Natural Equilibrium of Planet Earth

Requires to Reinvent Computing

Reiner Hartenstein

IEEE fellow

1

(ISCAS-2011)

Reiner

who is from SE (no free speech)

© 2010, [email protected] http://hartenstein.de

TU Kaiserslautern

2011,

(Preface) Without Computers?

2

Lufthansa anno 1960

(Business Information System)


TU Kaiserslautern

(Preface) very important future Applications

3

replacing bureaucracies by mass collaboration

http://www.macrowikinomics.comThe World Economic Forum:

other applications: see Cyber-Physical Systems

http://www.macrowikinomics.com/

Reiner

The Internet: new engine of innovation and wealth creation and even create public value - that radically drops collaboration costs. People everywhere are collaborating like never before, It’s a question of stagnation versus renewal. Atrophy versus renaissance. Peril versus promise.

Reiner

Maintaining the natural equilibrium of the planet earthalso to optimize all key issues given by the impact of the growing population of humans and their activities.Increasing Compute Capacity required

Reiner

Reiner07.06.2011the growing importance of connectivity for socio-economic interactions


TU Kaiserslautern

2011,

Preface

• Enormous Trouble in Computing:– Longterm Programming Crisis– Keynotes and Panel Discussions

booming– Excessive Power Consumption

4

Reiner

manycore and reconfigurable

Reiner

rapid growth of the internetgrowing no of wireless smart devices

Reiner

increasing number of important applications, not onlyaiming at the Natural Equilibrium of Planet Earth

Reiner

away from the Aristotelian world model, where von Neumann and Turing are the center of the world - over to a Copernican hetero world modela lot of voices believe in a some promising ideasmany voices are very pessimistic about the futuredevelopment of efficient parallelism

Reiner

contemporary computer systems are in an all-dominant programmability crisis. The progress of performance massively stalled because of the „programming wall“ caused by lacking scalability of parallelism and an ubiquitous programmer productivity gap.


TU Kaiserslautern

2011,

Outline (1)

•Energy consumption of Computers

•Toward Exascale Computing•The von Neumann Syndrome•We need to Reinvent Computing

•Conclusions5

Reiner

the IT area is responsible for the enormous growth of electricity consumption. In computer centers the energy expenses are growing 8 times faster than the hardware expense.


TU KaiserslauternBeyond peak oil

6

„6 more Saudi Arabias needed for demand predicted for 2030“

80% crude oil coming from

[Fatih Birol, Chief Economist IEA].

https://www.theoildrum.com/


TU Kaiserslautern

2011,

7 7

Saudi Arabia

© 2011 [email protected]


TU Kaiserslautern

2011,

How many more Saudi

Arabias needed?

Rio de Janeiro

8


TU Kaiserslautern

2011,

Power Consumption of the Internet

Power consumption by internet: x30 til 2030 if trends continueG. Fettweis, E. Zimmermann: ICT Energy Consumption - Trends and Challenges; WPMC'08, Lapland, Finland, 8 –11 Sep 2008

9

[Randy Katz: IEEE Spectrum, Febr. 2009]

more

than6

saudi

arabia !

Google Data Ccenter at Columbia River

soon 8 billion smart wireless devices

Reiner

Dallas at Columbia River

Reiner

hydroelectric power plants... amazon, MS, yahoo

Reiner

Reiner11.05.2011Already 3 years ago the carbon footprint of only the internet was higher than that of the worldwide air traffic. Under the growing oil price at declining production the rapidly growing energy consumption in all areas of computing will become unaffordable, probably within less than a decade.


TU Kaiserslautern

2011,

More Google Data Centers

10Google causing 2% electricity consumption worldwide ?

[datacenterknowledge.com]


10


TU Kaiserslautern

2011,

Electricity Bill: a Key Issue

„The possibility of computer equipment power consumption spiraling out of control could have

serious consequences for the overall affordability of computing.”

Patent for water-based data centers

Cost of a Google data center dominated only by monthly power bill

[L. A. Barroso, Google]

11

FERC

Google going to sell electricity

• Already in 2005, Google’s electricity bill higher than value of its equipment.

Reiner

Google asked the Federal Energy Regulatory Commission for the authority to sell electricityplanning several windmill parks, one of them off-shore at the east coast of the USA

Reiner

the ocean provides cooling & power (motion of surface) also note: the property tax !!!!

Reiner

inside Google the Cost of a data center is determined solely by the monthly power bill, not by the cost of hardware or maintenance


TU Kaiserslautern

2011,

12

The World's largest Data Center


Reiner

East CERMAK Chicago 1.1 mio square feet, until recently the world's largest data center


TU Kaiserslautern

2011,

Microsoft Data Center at Quincey

13



TU Kaiserslautern

2011,

About 2000 datacenters world-wide

14



TU Kaiserslautern

2011,

Outline (2)


•Toward Exascale Computing•The von Neumann syndrome•We need to Reinvent Computing

•Conclusions

15


TU Kaiserslautern

2011,

16year

relative performance

94 96 98 00 02 04 06 08 10 12 14 16 18 20 22 24 26 28 30

Multicore: Break-through or Breakdown?

x86parallelismvon-Neumann-only

much slower than Moore‘s law

beg

in o

f th

e

mult

icore

era

„forcing a historic transition to a parallel programming model yet to be invented“

David Callahan, Microsoft distinghuished endineer

perf

orm

anc

e gr

owth

need

ed

Reiner

& massive programmer productivity problems„Multicore shifts the burden of Performance from Chip Designer to Software Developers.“[J. Larus: Spending Moore's Dividend; C_ACM, May 2009]forcing a historic transition to a parallel programming model yet to be invented. [David Callahan, Microsoft distinguished engineer]

Reiner

System designers are actually experiencing performance drops when moving their single-core Extensive efforts are underway to help understand and resolve these performance-related issues, which also include issues associated with oversubscription, synchronization overhead, and intercore communications.


TU Kaiserslautern

2011,

„ intel has thrown a Hail Mary Pass“

DavePatterson

17

Reiner

*) a Hail Mary pass in American football is a forward pass made in desperation, with a very small chance of successThe Landscape of Parallel Computing Research: A View from Berkeley, 18 Dezember 2006“could reset µP HW & SW roadmaps for next 30 years”,„The entire IT Industry has bet on their future, that the problem of parallel prog'g will be solved."The stakes are high. When research will not find efficient parallel techniques, programming will become so difficult, that people will not have a benefit from the new hardware:from growth industry to replacement industryA year later: „I am still astonished about that"


TU Kaiserslautern

2011,

18

John Hennessy

„ … I would be panicking …“

Reiner

Hennessy: “… parallelism and ease of use of truly parallel computers: a problem that’s as hard as any that computer science has faced. … I would be panicked if I were in industry.” ---


TU Kaiserslautern

19

Exa-scale: (1018 computations/second) expected by 2018; [several sources]

Power estimated (single supercomputer):250 MW – 10 GW (2x NY City: 16 million people)

Exascale affordable ?

Reiner

more than 100,000 processor cores each, and distributed memory.

Reiner

They hope, that dozens of applications are inherently parallel enough to be laboriously decomposed, sliced and diced, for mapping onto HPC

Reiner

why does it take that long ?most apps written for single processorLarge applications only modestly scalable. >50% apps don’t scale beyond 8 cores, only 6% can exploit >128 PE cores, a tiny fraction 100,000 or more cores.


TU Kaiserslautern

20

In my opinion, the largest supercomputers at any time, including the first exaflops, should not be thought of as computers. …[Andrew Jones, vice president Numerical Algorithms Group]

Supercomputers: no Computers?


TU Kaiserslautern

21

…Their usage patterns and scientific impact are closer to major research facilities such as CERN, ITER, or Hubble.[Andrew Jones, vice president Numerical Algorithms Group]

no reason to solve the power problem ?

Supercomputers as Scientific Instruments


TU Kaiserslautern

2011,

CERN (1)

22

Reiner

Reiner10.05.2011an illustration of the monstrous financial projections of the numerical algorithms people


TU Kaiserslautern

2011,

CERN (2)23

Reiner

last image was only a peep hole projection compared to viewing the geographical size of CERN equipment


TU Kaiserslautern

2011,

Hubble 24

Reiner

Hubble: do not forget the immense expense to launch this heavy equipment into space and tolauch maintenance teams later on


TU Kaiserslautern

2011,

Learning how to go Exascale

CACHES 20111st International Workshop on Characterizing

Applications for Heterogeneous Exascale SystemsJune 4th, 2011, held in conjunction with ICS'2011

25th International Conference on SupercomputingMay 31 - June 4, 2011, Loews Ventana Canyon Resort, Tucson, Arizona

25

Reiner

ongoing research as well as future trends


TU Kaiserslautern

2011,

Outline (3)

•Energy consumption of Computers•Toward Exascale Computing•The von Neumann syndrome•We need to Reinvent Computing•Conclusions

26

Reiner

what is the reason of this ubiquitous disaster ?of the programming wall ???of high energy consumption? can we expect help from low power design?


TU Kaiserslautern

2011,

27

Reconfigurable Computing offers an overwhelming reduction of electricity consumption

Potential of RC

as well as massive speed-up factors …

explained by the von Neumann Syndrome

Reiner

… both by up to several orders of magnitude. Reconfigurable Computing can help to avoid becoming unaffordable .

Reiner

Reiner11.05.2011what is the reason of high energy consumption? the wong basic paradigm ?


TU Kaiserslautern

2011,

>15000>15000PISA project

28

FFTFFT100

Reed-Solomon Decoding

Reed-Solomon Decoding 2400

Viterbi DecodingViterbi

Decoding4001000

MACMAC

DSP and wireless

molecular dynamics

simulation

molecular dynamics

simulation88

BLASTBLAST52

protein identificatio

n

protein identificatio

n40

Smith-Waterman

pattern matching

Smith-Waterman

pattern matching

288

Bioinformatics

GRAPEGRAPE

2020AstrophysicsAstrophysics

SPIHT wavelet-based image compression


457

real-time face detection


60006000

video-rate stereo vision


900pattern

recognitionpattern

recognition 730

Image processing,Pattern matching,

Multimedia

3000CT imagingCT imaging

1

1000

1,000,000

Speedu

p-F

act

or

Speed-up factors are not newby avoiding the von Neumann paradigm

doub

led ev

. 4 m

onths

8723DNA seq.

100

10

10,000

100,000

1985 2000 201020051995


?

crypto

crypto1000

28500

DES breakingDES breaking

?


TU Kaiserslautern

2011,

Energy saving factors: ~10% of speedup

29

FFTFFT100

Reed-Solomon Decoding

Reed-Solomon Decoding 2400

Viterbi DecodingViterbi Decoding

4001000

MACMAC

DSP and wireless

molecular dynamics simulation

molecular dynamics simulation

88BLASTBLAST52

protein identification

protein identification

40

Smith-Waterman pattern matching

Smith-Waterman pattern matching

288

Bioinformatics

GRAPEGRAPE

2020AstrophysicsAstrophysics

cryptocrypto1000

28500DES breakingDES breaking

100

103

106

Sp

eed

up

-Fact

or

1995 2000 20102005

779

3439

http://hartenstein.de© 2011 [email protected]

doub

les ev

ery 4

months

Power save factors obtained



457



60006000video-rate

stereo vision


900pattern

recognitionpattern

recognition730

Image processing,Pattern matching,Multimedia

3000CT imagingCT imaging

8723DNA seq.

Reiner

Success with RC has been achieved in a variety of areas such as ......


TU Kaiserslautern

2011, 30

[Tarek El-Ghazawi et al.: IEEE COMPUTER, Febr. 2008]

Application . Speed-up factor

SavingsPower Cost Size

DNA and Protein sequencing 8723 779 22 253

DES breaking 28514 3439 96 1116much less

equipment needed

massively saving energy

RC*: Demonstrating the intensive Impact

SGI Altix 4700 with RC 100 RASC compared to Beowulf cluster

Tarek El-

Ghazawi

*) RC = Reconfigurable Computing

taxonomy of

HPRC

design

flows

12 %

9 %


TU Kaiserslautern

Drastically less Equipment needed

For instance: a hangar full of racks replaced byFor instance: a hangar full of racks replaced bya single rack without air conditioning

31

or ½ rack


TU Kaiserslautern

2011,

The Reconfigurability Paradox

• Routing congestion

32

• Lower clock speed

• Massive reconfigurability overhead

• Massive wiring overhead

Orders of magnitude better performance by a massively worse area-inefficient technology ?


TU Kaiserslautern

33

More power for creating foam

than to accelerate the vessel ?More power for creating foam

than to accelerate the vessel ?

33

because ofThe von Neumann Syndrome



TU Kaiserslautern

2011,

von Neumann Syndrome

34

Lambert M. Surhone, Mariam T. Tennoe,

Susan F. Hennessow (ed.): Von Neumann Syndrome; ßetascript publishing 2011


TU Kaiserslautern

2011,


von Neumann Model Critics35

“The von Neumann Syndrome”:[C.V. “RAM” Ramamoorthy 2007; UC Berkeley]

Nathan’s Law: Software is a gas. It expands to fill all its containers ...Nathan Myhrvold, Microsoft Ex-CTO

„even fills the internet“and the clouds

year

systemSLOC

(millions)

2001

Windows XP 40

2005

MAC OS X 10.4 86

2007

SAP Net Weaver 238

incompetent programmers

E. Dijkstra 1968; J. Backus 1978; Arvind , 1983; Peter G. Neumann 1985-2003; L. Savain 2006.

Critique of von Neumann is not new: N. N. 1995: THE STANDISH GROUP REPORTRobert N. Charette 2005: Why Software Fails; IEEE Spectrum

Anthony Berglas 2008: Why it is Important that Software Projects Fail

Software Desaster Reports:

Reiner Hartenstein

fundamentally wrong, since targetting data processing,not instruction streams

Reiner

due to Thomas Kuhnscientific communities behave by herd instinctSE reminds me to a buffalo herdThe term "Software" stands for the most harmful symptom of the vN Syndrome

Reiner

Reiner11.05.2011the reason for over-sized teams?The most often-overlooked risk in software engineering are incompetent programmers. We have a quality problem. One bad programmer can easily create two new jobs a year. [David Parnas]

Reiner

Reiner11.05.2011The Law of More: programmer productivity declines disproportionately with increasing parallelismHPC massive parallelism requires 10 – 30 professionalists in multi-disciplinary multi-insitutional teams for 5 - 10 years [Douglass Post, DoD HPCMP, panelist at SC07]


TU Kaiserslautern

2011,

All hardware but ALU is overhead: x20 inefficiency

36

x20 inefficiency: just one of

several overhead layers

[R. Hameed et al.: Understanding Sources of Inefficiency in General-Purpose Chips; 37th ISCA, June 19-23, 2010, St. Malo, France]

“GP Processors are inefficient”

(data cashe)

Reiner

recently much more literature about inefficient von Neumann paradigm

Reiner

But software is by orders of magnitude more inefficient


TU Kaiserslautern

2011,

„The Memory Wall“

60%/yr..

7%/year

Patterson’s Law:Processor-MemoryPerformance Gap:(grows 50% / year)

1

10

100

1000Performance

1980 1990 2000

DRAM

CPU

2008

>1000

coined by Sally McKee

The overwhealming problem is data moving complexity, not processor performance. Dr. Djordje Maric* (ETH Zurich),

37


TU Kaiserslautern

2011,

Through-Silicon-Via (TSV)

38

reduce power consumption by 75% [Wally Rh., Micro News 2/28/2011 ]

SIP multiple dicePoP Package on PackagePiP Package in PackageTSV Through silicon via

reducing the memory wall?

Reiner

Reiner11.05.20113D-IC packaging ideas have been around for more than 40 yearshopeful to lower the memory wall:well fits to Low Power Design Conferences

Reiner

not found in Low Power conferences


TU Kaiserslautern

Massive Overhead Phenomena

proportionate to the

number of processors

overproportionate to the number

of processors

39

overhead von Neumann machine

instruction fetch instruction streamstate address computation instruction streamdata address computation instruction streamdata meet PU + other overh. instruction streami / o to / from off-chip RAM instruction stream

Inter PU communication instruction streammessage passing overhead instruction streamtransactional memory overh. instruction streammultithreading overhead etc. instruction stream

Reiner

80ies: speed-up of up to 15,000


TU Kaiserslautern

40

von Neumann overhead vs. Reconfigurable Computing

overhead von Neumann machine datastream machine

instruction fetch instruction stream none*state address computation instruction stream none*data address computation instruction stream none*data meet PU + other overh. instruction stream none*i / o to / from off-chip RAM instruction stream none*Inter PU communication instruction stream none*message passing overhead instruction stream none*

transactional memory overh. instruction stream none*

multithreading overhead etc. instruction stream none*

*) c

onfig

ured

bef

ore

run

time

no inst

ruct

ion

fetc

h a

t ru

n t

ime

40


TU Kaiserslautern

2011,

Outline (4)


•Toward Exascale Computing•The von Neumann Syndrome•We need to Reinvent Computing

•Conclusions

41


TU Kaiserslautern

2011,

Putting Old Ideas Into Practice

42

Software Engineering http://www.acm.org/sigsoft/SEN/parnas.html SEN vol. 24 no. 3, May 1999

The biggest payoff will come from putting old ideas into practice (POIIP) and teaching people how to apply them properly. [David Parnas]

Reiner

which old ideas ?Most students who are studying computer science really want to study software engineering but they don't have that choice. There are very few programs that are designed as engineering programs but specialize in software. [David Parnas]

Reiner

Think there are fundamental design principles, for example structured programming principles, ... etc.

Reiner

Sakrileg: kostette ihn sein job ...

Reiner

Fundamental ideas never get out of date. However, some people would think of .... Turing machines and Goedel's theorem as fundamentals. ...... Those things are fundamental but they are also nearly irrelevant. [David Parnas]

Reiner

The most exciting/promising software engineering ideas or techniques are already here and have been here for years (or decades) but are not being used properly.


TU Kaiserslautern

2011,

Mike Flynn‘s Taxonomy

43

M. J. Flynn: “Very high-speed computing systems”; Proc. IEEE, Vol. 54, No. 12, pp. 1901–1909, Dec., 1966.

Reiner

Patterson: also, the old paradigm has changed. power was free, but transistors were expensive. Now power is expensive and transistors are free. chips were reliable internally. Now there's a huge increase in hard and soft errors on-chip. book: new business modelNICHT HIER !!!The old wisdom was that innovation came through compiler optimization. Now that innovation takes more than 10 years. We need a renaissance of performance, which can be had through using multiple processors. now the processor is the new transistor. We need to design complex systems with hundreds or thousands of processors on chip.The future is multicore and everything's going to be parallel.


TU Kaiserslautern

2011,

Diana‘s extended Taxonomy44© 2011 [email protected]

D. Göhringer, M. Hübner, T. Perschke, J. Becker: “A Taxonomy of Reconfigurable Single/Multi-Processor Systems-on-Chip”; International Journal of Reconfigurable Computing, Hindawi, Special Issue: Selected Papers from ReCoSoC 2008, 2009.

rSI: I can be reconfigured at run time: e. g. RISPrSD: can exchange data memory or datapathrSIrSD: both possible

4 x SISD:

4 x SIMD:

rMIrMD: supports both

I: instruction streamD: data stream

rMD: SIMD processors can exchange their data memories or reconfigure their datapaths

rSI: I can be reconfigured at run time: e. g. RISP

rSIrMD: can reconfigure both, D and Iat run time

rMI: MPSoCs w. reconfigurable I

4 x MIMD:

rMD: MPSoCs w. reconfigurable D

Reiner

teaching them properlyor generalization Aristoteles to Copernkcus


TU Kaiserslautern „But you can‘t implement decisions!“

45

S = R + (if C then A else B endif);

=1

+

ABR C

section of a very large pipe network:

decision

C. G. Bell et al: IEEE Trans-C21/5, May 1972W. A. Clark: 1967 SJCC, AFIPS Conf. Proc.

decision box turns into (de)multiplexer

**

Software to Configware Migration

0 1

(de)multiplexer:

B A

C

decision box:

C0 1

??

POIIP:

Reiner

the year 1971***:“That’s so simple! why did it take 30 years to find out?”

Reiner

DEC RTM-module system (before single-board computers came up

Reiner

time 2 space mapping


TU Kaiserslautern

POIIP: Loop to Pipe Mapping

46

(reconfigurable) DataPath Unit:

rDPUrDPUloop

body rDPUrDPU

rDPUrDPU

rDPUrDPU

Pipeline:

rDPUrDPUloop bod

y

loop:

complex loop bodynested loops

complex rDPU or pipe network inside rDPU

complex pipe

network

CPUCPU

MemoryMemory

Adder

Speaker

FMDemod

LPF1

Split

Gather

LPF2 LPF3

HPF1 HPF2 HPF3

Source:MIT

StreamIT

transport-triggered

Reiner

memory-cycle-hungry instruction streams (and data fetch/store)

Reiner

data streamsno instruction streams: no memory-cyclestransport-triggered


TU Kaiserslautern

POIIP: Loop to Pipe Mapping

47

(reconfigurable) DataPath Unit:

rDPUrDPUloop

body rDPUrDPU

rDPUrDPU

rDPUrDPU

Pipeline:

rDPUrDPUloop bod

y

loop:

complex loop bodynested loops

complex rDPU or pipe network inside rDPU

complex pipe

network

CPUCPU

MemoryMemory

Adder

Speaker

FMDemod

LPF1

Split

Gather

LPF2 LPF3

HPF1 HPF2 HPF3

Source:MIT

StreamIT

transport-triggered

on „platform FPGAs“

Reiner Hartenstein

i DO NOT WANT TO FRIGHTEN YOU


TU Kaiserslautern

48

Imperative Language Twins

language category Computer Languages Languages f. Anti Machine

both deterministic procedural sequencing: traceable, checkpointable

operation sequence driven by:

read next instruction, goto (instr. addr.),

jump (to instr. addr.), instr. loop, loop nesting

no parallel loops, escapes, instruction stream branching

read next data item, goto (data addr.),

jump (to data addr.), data loop, loop nesting, parallel loops, escapes, data stream branching

state register program counter data counter(s) address computation

massive memory cycle overhead overhead avoided

Instruction fetch memory cycle overhead overhead avoided parallel memory bank access interleaving only no restrictions

language features control flow + data manipulation

data streams only (no data manipulation)

very easy to learn

multipleGAGsmuch more

powerful

Flowware Languages Software Languages

much moresimple

von Neumann Languages

[COMP-EURO

’89]

Anti-machine:

MoPL: [FPL‘94, Prague] more simple

parallelism

solution

Reiner

easy to teach: simple machine principles

Reiner

JPEG zig-zag scan pattern


TU Kaiserslautern

2011,

A Heliocentric CS Model needed

49

PEProgram Engineering

The Generalization ofSoftware Engineering —

data streams*

*) do not confuse with „dataflow“!

Flowware Engineering

FE

auto-sequencing Memoryauto-sequencing Memory

asM asM

CEConfigware Engineering

structurespipe network modelpipe network model

rDPU rDPU reconfigurable-Data-Path- Unitreconfigurable-Data-Path- Unit

reconfigurable-Data-Path- Arrayreconfigurable-Data-Path- ArrayrDPA rDPA

instruction streams

SESoftware Engineering

CPU CPU

Reiner Hartenstein

program counter

Reiner Hartenstein

away from the CPU-centric antique CS world model

Reiner Hartenstein

data counter(s)

Reiner

Undiscovered, the datastream machine paradigm has been around for 3 decades: software used it indirectly via highly inefficient instruction streamsSystolic arrays popular since 1978Best and most universal DEF of "datastreams"


TU Kaiserslautern

50

program source

compilation result

Software instruction streams

Flowware data streams

Configware datapath structures configured

A Clean Terminology, please

Reiner

typical question: "microprogramming ?"


TU Kaiserslautern

2011,

Outline (5)

•Energy consumption of Computers•Toward Exascale Computing•The von Neumann Syndrome•We need to Reinvent Computing•Conclusions

51


TU Kaiserslautern

2011,

absurdely incomprehensible

abstractions

52

[For architecture design & debug] Concurrency models can operate at component architecture level rather than programming languages. [E. A. Lee]

[E. A. Lee: Are new languages necessary for multicore? 2007][E. A. Lee. The problem with threads. Computer, 2006.]

& Locality

Awareness

needed !!

We need model-based abstractions at algorithmic level

are the problem in „standard“ languages


TU Kaiserslautern

53

Higher Abstraction Levels

Efforts to extend standards-based, serial programming languages with features to describe parallel constructs are likely to fail.

Nick Tredennick:

Term Rewriting Systems (TRS) may raise the abstraction level up to math formulae

Mauricio Ayala-Rincón:

What’s more likely to succeed are languages that raise the level of abstraction in algorithm description

TRS: powerful for better language design and design space exploration

Reiner

e.g. TR-based design space explorer FELIX by Carlos Morra

Reiner

Levels of Parallelism Node Level: share memory and an interfaceSocket Level: Core Level: Vector Level: How wide vector or SIMD instrPipeline Level: instructions in completion at once.Instruction Level: RT Level (special architectures etc.)Logic Level (FPGAs)

Reiner

Development with VHDL is expensiveFPGAs Achilles’ Heel: long development time:Low level HDLs (VHDL/Verilog) are still dominantVery wide variety of tool flowsComplicated IP Core Scene: FPGA, eASIC, ASICThe Battle between FPGAs vs MPSoCis RTL vs. Software programming


TU Kaiserslautern

54

Conclusions

Twin Paradigm skills & basic hardware knowledge are essential qualifications for programmers.We urgently need a fundamental CS Education and Research Revolution for dual-rail-thinking

Since we‘ve to re-write software anyway we should do it twin-pardigm.We need a tool flow & education efforts supporting a twin-paradigm approach and locality awareness

Reiner

Reiner03.09.2008we cannot fully switch quickly to a disruptive paradigm shift. enormous burden by legacy software requires a mass movement...We need a twin paradigm approach, mainly based on computing wisdom mostly having been ignored by our curricula for decadescalls for a mass movement into a run-away computing revolution causing a quick, wide, and strong impact as known from the VLSI design revolution a la Mead & Conway.

Reiner

modern programmers:Yesterday’s Programmers: people understanding only software.embedded engineers: people who combine understanding of software and hardware.Programmers to-day: people who combine understanding of software and configwareunderstanding configware requires some understanding of hardware.combining understanding of software, configware, and hardware?

Reiner

5 bio $ marketestablished mainstream for > a decade

Reiner

Hennessy: “… parallelism and ease of use of truly parallel computers: a problem that’s as hard as any that computer science has faced. … I would be panicked if I were in industry.” --- Dave Patterson: need for a major, government-sponsored attack on the multicore challenge


TU Kaiserslautern We need „une' Levée en Masses“

55

We need „une' Levée en Masses“

We need „une' Levée en Masses“

55

Reiner Hartenstein

Hennessy: “… parallelism and ease of use of truly parallel computers: a problem that’s as hard as any that computer science has faced. … I would be panicked if I were in industry.” --- fund [only] three universities to get underway– Berkeley, Illinois, and Stanford - the need for a major, government-sponsored attack on the multicore challenge --- Since the first commercial computer in 1950, cost-performance of computing has improved by about 100 billion overall, using the rapidly increasing transistor speed the last 20 yearsarguing for a new Manhattan project: was TOP SECRET - academic career to support the war. Dave Patterson: need for a major, government-sponsored attack on the multicore challengeAlwyn Goodloe: I think “ease of use” [Hennessy] is really key in the effort. Most CIOs will probably say that recompiling the ole dusty deck is as much as they are willing to do. We probably need to say exactly what we mean by ease of use.


TU Kaiserslautern

2011,

56

Don‘t worry !Thank You very much !

too many panels and keynotes?

Reiner

Suddenly, All Computing Is Parallel: Seizing Opportunity Amid the Clamor - [Michael Wrinn]The proud era of von Neumann architecture passes into history - [Michael Wrinn]Foundational change will disrupt traditional habits throughout the discipline – [Michael Wrinn]


TU Kaiserslautern

2011,

END57


TU Kaiserslautern

time to space mapping

time domain: space domain:procedure domain structure domain

58

program loopn time steps, 1 CPU

pipeline1 time step, n DPUs

Bubble Sortn x k time steps,

1 „conditional swap“ unit

Shuffle Sortk time steps, n conditional swap“

time algorithm space algorithm

conditionalswap

xy

conditionalswap

conditionalswap

conditionalswap

conditionalswap

time algorithm space/time algorithm s

units

Reiner

Die Relativitätstheorie befasst sich mit der Struktur von Raum und Zeit.Die spezielle Relativitätstheorie befaßt sich mit der Relativität von Raum und Zeit .


TU Kaiserslautern

Architecture instead of synchro:

59

„Shuffle Sort“

conditionalswap

conditionalswap

conditionalswap

conditionalswap

modification: with shuffle-

function

conditionalswap

conditionalswap

conditionalswap

conditionalswap

conditionalswap

conditionalswap

swap

conditionalswap

conditional

direct time to space mappingaccessing conflicts

Better Architectureinstead of complex synchronisation: half he number of Blocks + up und down of data (shuffle function) – no von Neumann-syndrome !

Example

Reiner

Die Relativitätstheorie befasst sich mit der Struktur von Raum und Zeit.Die spezielle Relativitätstheorie befaßt sich mit der Relativität von Raum und Zeit .


TU Kaiserslautern

Understanding Complex Hetero Systems

60

Layers of Abstraction and Automatic Parallelization hide critical sources of, and limits to efficient parallel execution

Efficient Distribution of Tasks being memory limited

Internode Communications reduces Computational Efficiency

We must change how programmers think

essential: awareness of locality,

Focusing on memory mapping issues and transfer modes to detect overhead and bottlenecks

Understanding streams through complex fabrics needed

[Ed Lee]

Reiner

“Parallel programming: informal approaches are not working” [Mattson] Need a systematic approach from how programmers think” [Mattson]

Reiner

Architecture visible to the programmer:- The software programming model:Which hardware architecture parts are visible and under the programmer’s direct control ? - The RC programming model: how the programmer can control data transfers between FPGA, onboard memory, the microprocessor and its memory


TU Kaiserslautern

2011,

Vertical Disintegration

611960 200X

courtesy Manfred Glesner

Reiner

Complicated IP Core SceneTSMC IP cores collaboration program to improve soft IP readiness for EDAincluding Arteris, Atrenta, Cadence, Chips & Media, Imagination Technologies, Intrinsic-ID, MIPS Technologies, Sonics, Synopsys and Vivante.


TU Kaiserslautern

2011,

Market Complexity

62

Source: Gartner


TU Kaiserslautern

2011,

Taxonomy of Twin Paradigm Programming Flows (HPRC)

63

E. El-Araby et al.: Comparative Analysis of High Level Programming for Reconfigurable Computers: Methodology And Empirical Study; Proc. SPL2007, Mar del Plata, Argentina, Febr. 2007

[courtesy Richard Newton]

„The nroff of EDA“ [R. N.]


TU Kaiserslautern

2011,

HLL programming models

64


TU Kaiserslautern

Some hardware description languaqges

65

DeFactoGaladriel & NenyaMATCH


TU Kaiserslautern

Some programming languages

66


TU Kaiserslautern

Some languages for parallelism

67


TU Kaiserslautern

More Languages

Som

e f

un

ctio

nal la

ngu

ag

es

Som

e d

ata

stre

am

lan

gu

ag

es

68


TU Kaiserslautern

Why Computers are important

69 R. Rajkumar, I. Lee, L. Sha, J. Stankovic: Cyber-Physical Systems: The Next Computing Revolution; DAC 2010

Reiner

the growing importance of connectivity for socio-economic interactions


TU Kaiserslautern

2011,

Science alone ?

see the claims by Andrew Jones, …

70


TU Kaiserslautern

2011,

Mobile Communication2007 2014 2020

Worldwide radio base station sites* (millions)

3.3 7.6 11.2

Average power consumption per site (kW) 1.7 1.3 1.1

Total power consumption of all sites (TW) 5.6 10 12.5

Total global RAN energy consumption (TWh) 49 84 99

total # of subscriptions expected (billions) 6 9

Broadband subscriptions expected (billions) 2

Video streams (%) 66 90

Share of mobile data in total mobile traffic (%)

37.5 98 99.6

71

A. Fehske, J. Malmodin, G. Biczók, G. Fettweis: The Global Footprint of Mobile Communications – The Ecological and Economic Perspective; IEEE Communications Magazine, Aug 2011

*) all standards

The data transmission speed growth by a factor of ten every five years (cellular, local + personal area networks),

Technologies to reduce energy consumption are a key

enabler

Reiner

radio access networks (RANs)


TU Kaiserslautern

2011,

Undersea Cable

72

Google: 9,620km submarine cable Japan-US; 1st use Febr 21, 2011Five fiber pairs deliver up to 4.8 Terabits per second (Tbps)

>100 kilometers between repeaters

wavelength-division multiplexing dramatically increases fiber capacity.

repeater laser power consumption <25 W

power consumption of fabrication and cable layer ships much higher

multiple (e.g. 5) pairs of fibers: each pair has one fiber in each direction

<1000 repeaters: <25 kW


TU Kaiserslautern

2011,

73 73

Saudi Arabia



TU Kaiserslautern

2011,

How many more Saudi

Arabias needed?

Rio de Janeiro

74


TU Kaiserslautern

2011,

Undersea Cable

75

Google: 9,620km submarine cable Japan-US; 1st use Febr 21, 2011Five fiber pairs deliver up to 4.8 Terabits per second (Tbps)

>100 kilometers between repeaters

wavelength-division multiplexing dramatically increases fiber capacity.

repeater laser power consumption <25 W

power consumption of fabrication and cable layer ships much higher

multiple (e.g. 5) pairs of fibers: each pair has one fiber in each direction

<1000 repeaters: <25 kW

aiming at the natural equilibrium of planet earth requires to reinvent computing reiner hartenstein...

Documents

tu kaiserslautern preface

scientific instruments

mass collaboration http

electricity consumption

internet power consumption

excessive power consumption

google data centers

computing conclusions