1. Parallel Processing: Architecture Overview Subject Code: 433-498 Rajkumar Buyya Grid Computing and Distributed Systems (GRIDS) Lab. The University of Melbourne Melbourne, Australia www.gridbus.org WW Grid

2. Overview of the Talk  Why Parallel Processing ?  Parallel Hardwares  Parallel Operating Systems  Parallel Programming Paradigms  Grand Challenges

3. PPPPPP  Microkernel Multi-Processor Computing System Threads Interface Hardware Operating System Process Processor Thread P P Applications Computing Elements Programming paradigms

4. Architectures System Software/Compiler Applications P.S.Es Architectures System Software Applications P.S.Es Sequential Era Parallel Era 1940 50 60 70 80 90 2000 2030 Two Eras of Computing Commercialization R & D Commodity

5. History of Parallel Processing PP can be traced to a tablet dated around 100 BC. Tablet has 3 calculating positions. Infer that multiple positions: Reliability/ Speed

6. Motivating factors  Just as we learned to fly, not by constructing a machine that flaps its wings like birds, but by applying aerodynamics principles demonstrated by the nature...  We modeled PP after those of biological species.

7. ß Aggregated speed with which complex calculations carried out by neurons-individual response is slow (ms) – demonstrate feasibility of PP Motivating Factors

8. Why Parallel Processing? Ø Computation requirements are ever increasing -- visualization, distributed databases, simulations, scientific prediction (earthquake), etc. Ø Sequential architectures reaching physical limitation (speed of light, thermodynamics)

9. Age Growth 5 10 15 20 25 30 35 40 45.... Human Architecture! Growth Performance Vertical Horizontal

10. No. of Processors C.P.I. 1 2.... Computational Power Improvement Multiprocessor Uniprocessor

11. Ø The Tech. of PP is mature and can be exploited commercially; significant R & D work on development of tools & environment. Ø Significant development in Networking technology is paving a way for heterogeneous computing. Why Parallel Processing?

12. Ø Hardware improvements like Pipelining, Superscalar, etc., are non- scalable and requires sophisticated Compiler Technology. Ø Vector Processing works well for certain kind of problems. Why Parallel Processing?

13. Parallel Program has & needs... ä Multiple “processes” active simultaneously solving a given problem, general multiple processors. ä Communication and synchronization of its processes (forms the core of parallel programming efforts).

14. Processing Elements Architecture

15. ä Simple classification by Flynn: (No. of instruction and data streams) > SISD - conventional > SIMD - data parallel, vector computing > MISD - systolic arrays > MIMD - very general, multiple approaches. ä Current focus is on MIMD model, using general purpose processors. (No shared memory) Processing Elements

16. SISD : A Conventional Computer  Speed is limited by the rate at which computer can transfer information internally. Processor Data Input Data Output Instructions Ex:PC, Macintosh, Workstations

17. The MISD Architecture  More of an intellectual exercise than a practicle configuration. Few built, but commercially not available Data Input Stream Data Output Stream Processor A Processor B Processor C Instruction Stream A Instruction Stream B Instruction Stream C

18. SIMD Architecture Ex: CRAY machine vector processing, Thinking machine cm* Intel MMX (multimedia support) C i <= A i * B i Instruction Stream Processor A Processor B Processor C Data Input stream A Data Input stream B Data Input stream C Data Output stream A Data Output stream B Data Output stream C

19. Unlike SISD, MISD, MIMD computer works asynchronously. Shared memory (tightly coupled) MIMD Distributed memory (loosely coupled) MIMD MIMD Architecture Processor A Processor B Processor C Data Input stream A Data Input stream B Data Input stream C Data Output stream A Data Output stream B Data Output stream C Instruction Stream A Instruction Stream B Instruction Stream C

20. MEMORYMEMORY BUSBUS Shared Memory MIMD machine Comm: Source PE writes data to GM & destination retrieves it  Easy to build, conventional OSes of SISD can be easily be ported  Limitation : reliability & expandibility. A memory component or any processor failure affects the whole system.  Increase of processors leads to memory contention. Ex. : Silicon graphics supercomputers.... MEMORYMEMORY BUSBUS Global Memory System Processor A Processor A Processor B Processor B Processor C Processor C MEMORYMEMORY BUSBUS

21. MEMORYMEMORY BUSBUS Distributed Memory MIMD l Communication : IPC on High Speed Network. l Network can be configured to... Tree, Mesh, Cube, etc. l Unlike Shared MIMD  easily/ readily expandable  Highly reliable (any CPU failure does not affect the whole system) Processor A Processor A Processor B Processor B Processor C Processor C MEMORYMEMORY BUSBUS MEMORYMEMORY BUSBUS Memory System A Memory System A Memory System B Memory System B Memory System C Memory System C IPC channel IPC channel

22. Laws of caution..... l Speed of computers is proportional to the square of their cost. i.e. cost = Speed Speedup by a parallel computer increases as the logarithm of the number of processors. Speedup = log2(no. of processors) S P log 2 P C S (speed = cost 2 )

23. Caution.... ã Very fast development in PP and related area have blurred concept boundaries, causing lot of terminological confusion : concurrent computing/ programming, parallel computing/ processing, multiprocessing, distributed computing, etc.

24. It ’ s hard to imagine a field that changes as rapidly as computing.

25. Computer Science is Immature Science. (lack of standard taxonomy, terminologies) Caution....

26. ã Even well-defined distinctions like shared memory and distributed memory are merging due to new advances in technolgy. ã Good environments for developments and debugging are yet to emerge. Caution....

27. ã There is no strict delimiters for contributors to the area of parallel processing : CA,OS, HLLs, databases, computer networks, all have a role to play. Ü This makes it a Hot Topic of Research Caution....

28. Operating Systems for High Performance Computing

29. Types of Parallel Systems ä Shared Memory Parallel ç Smallest extension to existing systems ç Program conversion is incremental ä Distributed Memory Parallel ç Completely new systems ç Programs must be reconstructed ä Clusters ç Slow communication form of Distributed

30. Operating Systems for PP MPP systems having thousands of processors requires OS radically different fromcurrent ones. Every CPU needs OS : to manage its resources to hide its details Traditional systems are heavy, complex and not suitable for MPP

31. 2 Frame work that unifies features, services and tasks performed 2 Three approaches to building OS.... * Monolithic OS * Layered OS * Microkernel based OS Client server OS Suitable for MPP systems 2 Simplicity, flexibility and high performance are crucial for OS. Operating System Models

32. Application Programs Application Programs System Services Hardware User Mode Kernel Mode Monolithic Operating System c Better application Performance c Difficult to extend Ex: MS-DOS

33. Layered OS lEasier to enhance lEach layer of code access lower level interface lLow-application performance Application Programs Application Programs System Services User Mode Kernel Mode Memory & I/O Device Mgmt Hardware Process Schedule Application Programs Application Programs Ex : UNIX

34. Traditional OS OS Designer OS Hardware User Mode Kernel Mode Application Programs Application Programs Application Programs Application Programs

35. New trend in OS design User Mode Kernel Mode Hardware Microkernel Servers Application Programs Application Programs Application Programs Application Programs

36. Microkernel/Client Server OS (for MPP Systems) lTiny OS kernel providing basic primitive (process, memory, IPC) lTraditional services becomes subsystems lMonolithic Application Perf. Competence lOS = Microkernel + User Subsystems Client Application Client Application Thread lib. Thread lib. File Server File Server Network Server Network Server Display Server Display Server Microkernel Hardware User Kernel Send Reply Ex: Mach, PARAS, Chorus, etc.

37. Few Popular Microkernel Systems, MACH, CMU, PARAS, C-DAC, Chorus, QNX,, (Windows)