The Tunnel Vision Syndrome: Massively Delaying Progress

Transcription

1 VIPSI-2012 MONTENEGRO Hotel Splendid in Becici Dec 31, 2012 to Jan 1, 2013 The George Washington University, 5-7 June 2013, Washington, DC Reiner Hartenstein IEEE fellow FPL fellow SDPS fellow The Tunnel Vision Syndrome: Massively Delaying Progress Kaiserslautern University of Technology 1

2 Outline (1) Preface History of Computing The Systolic Array The Kress Array The Xputer Paradigm The Twin Paradigm Approach More Tunnel Vision We must Reinvent Computing Conclusions 2

3 mini computers: VAX-11/750 quasi-standard around 1980 (personal experience:) UC Berkeley (visiting professor) my CS department at Kaiserslautern my Xputer lab at Kaiserslautern E.I.S. project (my M-&-C-action) NATO ASI, Urbino 1981 my von Neumann syndrome experience also here 3

4 Too many terminals Paolo Antognetti Data Input Speed-up orders of magnitude by Orders speed-up by using of Magnitude museum equipment! Donald O. Pederson I used this Museum Equipment NATO ASI on VLSI 1981, SOGESTA, Urbino, Italy, 4

5 The Beauty and the Joy of Computing??? Lambert M. Surhone, Mariam T. Tennoe, Susan F. Hennessow (ed.): Von Neumann Syndrome; ßetascript publishing

6 [1980] Education Revolution: the M-&-C VLSI Design Revolution Das E.I.S.-Projekt: The by far most effective project in the history of modern computer science 2012, 6

7 Dual paradigm mind set: an old hat - but ignored time to space mapping: procedural to structural: loop to pipe mapping 1971: why did it take 25 years to find out? token bit evoke FF 1967: W. A. Clark: Macromodular Computer Systems; 1967 SJCC, AFIPS Conf. Proc. C. G. Bell et al: The Description and Use of Register- Transfer Modules (RTM's); IEEE Trans-C21/5, May FF FF

8 Diana Göhringer s Ph.D.thesis A huge design space Mike Flynn s taxonomy Programmability crisis solution impossible without mastering the entire design space extending Flynn s taxonomy by going heterogeneous: Instruction vs. Data Single vs. Multiple Diana s Taxonomy Reiner s Taxonomy reconfigurable or not 8 datastream-based (anti-machine) noi versus SI or MI

9 The tunnel vision of the pre-manycore age 9

10 Heterogeneous Taxonomy The extension into this huge design space is mainly ignored by curricula and even by most R&D scenes 10

11 fragmentation tall thin man coherence traditional division of specialization: submit Register Transfer (RT) level submit submit submit submit Application level Logic level Switching level Circuit level Layout level in-house technology width of specialization Education Revolution: the M-&-C Design Revolution reject reject reject reject reject 11 The Mead-&-Conway strategy: Das E.I.S.-Projekt: Clearing out & intuitive models Removal of the education dilemma Application Silicon Foundry reduced width of specialization

12 Less Abstraction Layers? Removing Abstraction Layers hides critical sources of efficiency limits. Hides issues to detect overhead and bottlenecks We must change how programmers think, also by.. fusion of abstraction layers: challenging educators opening the borders between paradigm domains 12

13 13 The Tunnel Vision of Language Designers in teaching to students Absurdely incomprehensible abstractions are the problem in standard languages The High Cost of Movement of Data and Instructions Concurrency models should operate at component architecture level rather than programming languages. [E. A. Lee] 13 [E. A. Lee: Are new languages necessary for multicore? 2007] [E. A. Lee. The problem with threads. Computer, 2006.]

14 tall thin man coherence tall thin woman coherence even more More Connected Thinking traditional division of specialization: application level global and local memory distributed memory pipe-networks arbiter off-chip communication Multiplexing Interconnect at all levels distribution inter-processor NoC Register Transfer (RT) level Logic level Switching level Circuit level Layout level 14 The new strategy: heterogeneous platforms

15 The End of Evolution The end of evolutionary extension of current models Disruptive research is required in programmability Fundamental programming issues have to be revisited An entirely new software stack is needed Rethink all disciplines from circuit design and test, up to architecture, system design, storage behavior, compilers, run time systems, operating systems, and programming. We need a much deeper integration between applications and data inside all kinds of memory 15

16 Conclusions We need by orders of magnitude more parallelism and power-efficiency A reductionist attitude of most R&D areas massively delays the solution of urgent problems Disruptive research is urgently required and fundamental issues have to be revisited We must disruptively reivent CS education We must overcome the von Neumann syndrome and the widespread Tunnel Vision dementia 16