System 2020:Research Grand Challenges in Computer ArchitectureMary Jane IrwinPenn State University

John ShenIntel

What is the next big thing ?MainframesMinisWorkstationsPCs???

What are the mega trends ?Wired Wireless Telecommunication Internet/ComputingPatch-work Wireless Blanket WirelessPersonal Computer Mobile Computer Persistent/Transparent ComputerEmbedded vs. High-end Convergence?Client vs. Server Convergence?

And anticipated usage models ?Human-centric: Intelligent spacesAt home, work, school, leisure, Active displays, sensory devices, immersive experiences, Personal agentsFeature rich gadgets, real-time informationHighly mobile - roam seamlessly from space to spaceInfrastructure-centric: Traditional server farms and data centersVery large scale information fusion, storage, analysisSupporting an enormous number of roaming servers Fabric for supporting human-centric usesProactively pushing information to roaming agentsCommunication and synchronization between spaces

The computing paradigm ala Google

The computing paradigm ala Nokia

CRA System 2020 WorkshopLogisticsDecember 4 (pm), 5, 6, and 7 (am), 2005 at the Seascape Resort, Monterey Bay55 participants (15 industry, 33 academia, 7 CRA and NSF)Structuretwo bookend keynotes: Shekhar Borkar (Intel) on Microarchitecture Challenges for 2015 and Jim Larus (MSR) on Software Challenges in the Nanoscale Technologies. one industry panel: Bob Colwell (Consultant), Chuck Moore (AMD), Ravi Nair (IBM), Justin Rattner (Intel), and Steve Scott (Cray)rest was brainstorming with afternoon report out plenariesManaged by CRA, funded by NSF

What are the components of a GC?A grand scale problem that will require at least a decade of concentrated research to make substantive progressthat has a measurable outcomes/milestones,that will excite and engage the computer architecture research community,and that is deserving of considerable investment by funders because it will materially advance the capabilities and conduct of society.

1W Featherweight SupercomputerFor the goal of 1TOP/W will need up to 10,000X improvement in EPO (energy/op)1TOP/W = .001 nJ/op vs. todays ~10nJ/op

Architects are already engaged Societal impacts and fundingSocietal impacts are clear and compelling: pervasive intelligent sensors, embedded supercomputing appliances, . . .Funding investments ?

Featherweight Challengespower/energy reduction issuesdynamic and leakage, HW/SW mode controls, . . .performance improvement issuesCMPs & SMT, heterogeneous cores, programmable accelerators, eDRAMs, NoCs, . . .technology concerns (65nm45nm32nm) process variation, transient/permanent faults, advanced packaging (SoC MCP 3D), . . .design and fabrication concernsdesign time & tools, verification & test, fab costs. . .programmability . . .

Popular Parallel Programming (P3)Software and architecture support that makes parallel programming easyIf 2X per two year perf. gains continue, will soon have 1000-way chip-level parallelism

Architects are becoming engaged but cant do the job aloneNeed compiler, prog. languages, OS & application developersSocietal impacts and fundingA necessary enabling technology for future chips (e.g., the 1W Featherweight Supercomputer)Funding investments ?

P3 Challengesnew languages/models/compiler issuesthat are correct, efficient, scalable, portable, . . .that require minimal exposure of the programmer to low-level detailsand that support multi-modal parallelismdata-parallel, embarrassingly parallel, irregularly parallelmicroarchitecture support issueslightweight thread/process launch, communication and synchronization, monitoring for reliability and thermal hot spots with dynamic adaptation, . . .development supportbenchmarks, prototyping platforms, tools for debugging, performance tuning, . . .

Dependable Systems1. Self-healing hardware and software systems that you can trust your life on2x improvement in mean work-to-failure per generationwhile reducing the cost of ownership and vendor costs for liability/repair

2. Architects already engaged but cant do the job aloneA system stack problem devices, circuits, archs, languages, OS, applications, dependability analysts3. Societal impacts and fundingThe s/w problem alone is ~ 0.6% GDP of the USFunding investments ?

Dependable Systems ChallengesHost of hardware reliability problemsTransient SEU, temperature/process variations, Permanent aging, TDDB, NBTI, EM, Software reliability issuesIncreasing security issues Dynamically adapt to system constraints of reliability, security, performance, power Architects can provide low cost solutionsWorkload-aware, selective, fast, adaptiveBring dependability to h/w-s/w interfaceIntegrated cross-layer solution from devices to applications

New Computing ModelsBeyond the stored program architecturedata flow? neural network?Expanding the box for architectsneuroscientists, biologists, chemists, . . .Societal impacts and fundingNeuro-prosthetics, telepathy, . . .Funding investments ?

Brain ChallengesHigh risk but high payoffNeuroscientists are a long way from unraveling the mysteries of the neocortexTake partial steps augment certain brain functions (hearing for the deaf, vision for the blind, mobility for the quadrapeligic), Take advantage of emerging technologiesHeterogeneous systems: silicon + nanosensors and actuators, emerging nanotechnologies (CNT, QCAs, quantum, . . .)

Watch for the final reporthttp://www.cra.org/Activities/grand.challenges/architecture/home.html

And check out the reports from the previous Grand Challenges conferenceshttp://www.cra.org/grand.challenges/

Google: [NYT, June 8, 2006]2001 (March)8,000 servers70M pages/day2003100,000 servers2006450,000 servers20101,000,000 servers

Microsoft200,000 servers (2006) 800,000 servers (2011)

Yahoo Not just Nokia, but also Samsung, Motorola, etc.

Land a man on the moon and bring him back safely by the end of the decade. J.F. Kennedy