To Boldly Go. Emergenet, York, 20 th. April, (an occam-π mission on engineering emergence)

Transcription

1 To Boldly Go (an occam-π mission on engineering emergence) Peter Welch, Kurt Wallnau, Adam Sampson, Mark Klein 1 School of Computing, University of Kent Software Engineering Institute, Carnegie-Mellon University Emergenet, York, 20 th. April, Apr-10 Copyright P.H.Welch 1

2 To Boldly Go: an occam-π Mission A thesis, boids and a demo Process architecture and boids Observations of emergence Summary and Conclusions 18-Apr-10 Copyright P.H.Welch 2

3 Engineering Emergence THESIS Some future systems will be too complex to design and implement explicitly. Instead, we will have to learn to engineer the desired behaviours implicitly. We will do this through the discovery and programming of simple rules of behaviour, applied to a mass of dynamically configured and interacting components, from which desired complex behaviours emerge 18-Apr-10 Copyright P.H.Welch 3

4 Engineering Emergence THESIS Some future systems will be too complex to design and implement explicitly. Instead, we will have to learn to engineer the desired behaviours implicitly. The components individually will be simple,, showing not a hint of the complex behaviours that can emerge when a lot of them get together 18-Apr-10 Copyright P.H.Welch 4

5 Engineering Emergence Examples? Mechanisms design (game theory, micro-economics) Rational actors have local, private information Emergent: optimal allocation of scarce resources Optimal decisions rely on truth revelation SEI Computational Mechanism Design 18-Apr-10 Copyright P.H.Welch 5

6 Engineering Emergence Examples? Swarming behaviour (flocks, wasp colony behavior) Autonomous (non-rational) actors, local interactions only Emergent: swarm behavior UAV swarms and autonomous robots UAV SWARM HEALTH MANAGEMENT Aerospace Controls Laboratory, MIT (see 18-Apr-10 Copyright P.H.Welch 6

7 Engineering Emergence Examples? Social communication (gossip, epidemic algorithms) Large, ad hoc, dynamic networks Emergent: minimum power to achieve eventual consistency Low power, low reliability sensors and data propagation Self-regulating sensor networks, Trickle algorithm, Stanford (see August 08 issue of CACM) 18-Apr-10 Copyright P.H.Welch 7

8 Engineering Emergence Case study Boids: avoid collisions, match vector with those of birds is sight, head for the centre of mass of birds in sight, take fright if a hoik is spotted, be attracted by foid, Emergent behaviours: flocking, squabbling, migration waves, panic scattering, orbiting points of attraction (if only a small group),, feeding frenzy (if a large enough flock),, turbulence, maze solving, 18-Apr-10 Copyright P.H.Welch 8

9 demo occoids cylons 18-Apr-10 Copyright P.H.Welch 9

10 To Boldly Go: an occam-π Mission A thesis, boids and a demo Process architecture and boids Observations of emergence Summary and Conclusions 18-Apr-10 Copyright P.H.Welch 10

11 Lightweight Communicating Processes Fine-grained Massively parallel (zillions) Process-oriented oriented This is the way of the world... Processes, networks, networks-within within-networksnetworks Channel (reader-writer) synchronisation Barrier (multiway synchronisation) Ever-changing network topologies Dynamic birth, re-connections, death Mobile channels and processes CSP / occam-π π-calc / occam-π Mobile process location and neighbour awareness 18-Apr-10 Copyright P.H.Welch 11

12 foo bar m e r g e server (a) a network of three processes, connected by four internal (hidden) and three external channels. (b) three processes sharing the client end of a channel bundle to a server process. p (0)... p (n-1)... s (0) s (7) (c) three processes sharing the client end of a channel bundle to a bank of servers sharing the other end. (d) n processes enrolled on a shared barrier (any process synchronising must wait for all to synchronise). 18-Apr-10 Copyright P.H.Welch 12

13 Location (Neighbourhood) Awareness The Matrix Mobile Agents 18-Apr-10 Copyright P.H.Welch 13

14 Location (Neighbourhood) Awareness 18-Apr-10 Copyright P.H.Welch 14

15 Location (Neighbourhood) Awareness 18-Apr-10 Copyright P.H.Welch 15

16 Location (Neighbourhood) Awareness 18-Apr-10 Copyright P.H.Welch 16

17 occam-π Boids Model Each server is responsible for its own region of space A region may hold many birds or none Each bird is in only one region at a time but can consult with its immediately neighbouring regions 18-Apr-10 Copyright P.H.Welch 17

18 Location (Neighbourhood) Awareness So, not this 18-Apr-10 Copyright P.H.Welch 18

19 Location (Neighbourhood) Awareness but this 18-Apr-10 Copyright P.H.Welch 19

20 Each bird registers its state (position, vector, alarm state, colour, etc.) to the server for its region 18-Apr-10 Copyright P.H.Welch 20

21 (0.000, 0.000) (1.000, 0.000) (0.318, 0.788) (0.000, 1.000) (1.000, 1.000) Each bird knows its position relative to its current region of space it doesn t t know which region that is 18-Apr-10 Copyright P.H.Welch 21

22 Birds have a maximum range of vision (up to a radius of 1) 18-Apr-10 Copyright P.H.Welch 22

23 Birds have a maximum range of vision (up to a radius of 1) so may need to consult up to 4 servers 18-Apr-10 Copyright P.H.Welch 23

24 Birds also have a restricted angle of vision in this case to 300º (i.e. missing 60º rear view) 18-Apr-10 Copyright P.H.Welch 24

25 occam-π Boids Model A bird process follows a general pattern for mobile agents It has a pilot sub-process, responsible for dealing with the servers in its immediate neighbourhood and, when necessary, moving between them. The pilot is the eyes and wings of the bird It has brain sub-processes, receiving vision information from the pilot and computing wing muscle forces back to the pilot 18-Apr-10 Copyright P.H.Welch 25

26 occam-π Boids Model occoid Two-way channel bundles to own regional server + eight immediate neighbours 18-Apr-10 Copyright P.H.Welch 26

27 occam-π Boids Model occoid collision avoidance Σ center of mass filter.vision pilot match vector 18-Apr-10 Copyright P.H.Welch 27

28 occam-π Boids Model collision avoidance occoid center of mass filter.vision pilot match vector Actual Actual Actual 18-Apr-10 Copyright P.H.Welch 28

29 occam-π Boids Model A bird process follows a general pattern for mobile agents The birds are kept in step with each other (and with a visual renderer process) by barrier syncs which also provides a model of time. The pilot process does this 18-Apr-10 Copyright P.H.Welch 29

30 Barrier Synchronisation The occam-π BARRIER type corresponds to a multiway CSP event, though some higher level design patterns (such as resignation) have been built in. worker (0) worker (1) worker (n-1) b Basic CSP semantics apply. When a process synchronises on a barrier, it blocks until all other processes enrolled on the barrier have also synchronised. Once the barrier has completed (i.e. all enrolled processes have synchronised), all blocked processes are rescheduled for execution. 18-Apr-10 Copyright P.H.Welch 30

31 occam-π Boids Model collision avoidance occoid center of mass filter.vision pilot match vector Actual Actual Actual 18-Apr-10 Copyright P.H.Welch 31

32 occam-π Boids Model A bird process follows a general pattern for mobile agents The birds are kept in step with each other (and with a visual renderer process) by barrier syncs which also provides a model of time. The pilot process does this all see WHILE alive a consistent SEQ view and, possibly, SYNC tick move... observe local neighbourhood SYNC tick... change local neighbourhood 18-Apr-10 Copyright P.H.Welch 32

33 occam-π Boids Model A regional server process holds a dynamic array of all visiting birds It supplies this information to all observers: the birds, the process doing the rendering and, in future, live hawks, food, etc. These server processes do not sync on the barrier they have no need keep note of time or keep in step with the birds. 18-Apr-10 Copyright P.H.Welch 33

34 To Boldly Go: an occam-π Mission A thesis, boids and a demo Process architecture and boids Observations of emergence Summary and Conclusions 18-Apr-10 Copyright P.H.Welch 34

35 Engineering Emergence Case study reminder Boids: avoid collisions, match vector with those of birds is sight, head for the centre of mass of birds in sight, take fright if a hoik is spotted, be attracted by foid, Emergent behaviours: flocking, squabbling, migration waves, panic scattering, orbiting points of attraction (if only a small group),, feeding frenzy (if a large enough flock),, turbulence, maze solving, 18-Apr-10 Copyright P.H.Welch 35

36 Engineering Emergence Case study reminder Almost all processes have been described (5x800) bird processes, (8x5) regional servers.. There are only 4 others (for visual rendering and keyboard input). Emergent behaviours: flocking, squabbling, migration waves, panic scattering, orbiting points of attraction (if only a small group),, feeding frenzy (if a large enough flock),, turbulence, maze solving, 18-Apr-10 Copyright P.H.Welch 36

37 Engineering Emergence Case study reminder There is nothing in the design or programming dealing with flocking, scattering, orbiting, feeding frenzies, migration waves, turbulent flow or solving mazes! Emergent behaviours: flocking, squabbling, migration waves, panic scattering, orbiting points of attraction (if only a small group),, feeding frenzy (if a large enough flock),, turbulence, maze solving, 18-Apr-10 Copyright P.H.Welch 37

38 Engineering Emergence Case study reminder We don t t like the scattering we would prefer the flock to maintain cohesion when danger is spotted and turn-as-one away from it but what are the rules for engineering this behaviour? There is no concept of flock (for example) in the design so there is nothing to program directly. The panic signal propagates fast across a flock but the birds don t t have the right rules for the right response to emerge. Any ideas? 18-Apr-10 Copyright P.H.Welch 38

39 Engineering Emergence Scheduling dyamics reminder The network topology changes all the time as the birds move The computational loading on each bird and each server varies dynamically and cannot be predicted in advance Nevertheless, the occam-pi kernel (CCSP) does a good job of very lightweight load balancing across all the cores (that we have right now!) 18-Apr-10 Copyright P.H.Welch 39

40 To Boldly Go: an occam-π Mission A thesis, boids and a demo Process architecture and boids Observations of emergence Summary and Conclusions 18-Apr-10 Copyright P.H.Welch 40

41 To Boldly Go Summary We have described an architecture for the intentional emergence of complex systems behaviour. Processes (mobile, communicating and lightweight) ) are good candidates for supporting such an architecture. occam-π provides this computaional model and scales well across both shared and distributed memory. Engineering the desired behaviour is indirect.. We need to discover simple low-level level rules for pieces that we can program and, then, run masses of them. For complex systems, there will be no high-level components that directly work the behaviour we want. 18-Apr-10 Copyright P.H.Welch 41

42 To Boldly Go Summary It s It s It s programming, programming, programming, Jim, Jim, Jim, but but but not not not as as as we we we know know know it it it 18-Apr-10 Copyright P.H.Welch 42

43 Scientific Instruments Scientist / Engineer Instrument Complex System 18-Apr-10 Copyright P.H.Welch 43

44 Simulation as a Scientific Instrument * Scientist / Engineer Simulation Complex System Modelling Simulation Validation cycle Research and discovery of low- level processes from which observed complex behaviours emerge. * Simulation as an experimental design process for emergent systems, Andrews-Stepney-Winfield 18-Apr-10 Copyright P.H.Welch 44

45 Discovering Unexpected Relations between Phenomena Starlings Scientist / Engineer Simulation Modelling Simulation Validation cycle Research and discovery of low- level processes from which observed complex behaviours emerge. 18-Apr-10 Copyright P.H.Welch 45

46 Discovering Unexpected Relations between Phenomena Starlings Scientist / Engineer Simulation Modelling Simulation Validation cycle Research and discovery of low- level processes from which observed complex behaviours emerge. 18-Apr-10 Copyright P.H.Welch 46

47 Discovering Unexpected Relations between Phenomena Starlings Scientist / Engineer Simulation Modelling Simulation Validation cycle Research and discovery of low- level processes from which observed complex behaviours emerge. 18-Apr-10 Copyright P.H.Welch 47

48 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Modelling Simulation Validation cycle Research and discovery of low- level processes from which observed complex behaviours emerge. 18-Apr-10 Copyright P.H.Welch 48

49 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Modelling Simulation Validation cycle Research and discovery of low- level processes from which observed complex behaviours emerge. 18-Apr-10 Copyright P.H.Welch 49

50 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Modelling Simulation Validation cycle Research and discovery of low- level processes from which observed complex behaviours emerge. 18-Apr-10 Copyright P.H.Welch 50

51 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Computer modelling and simulaton can show unexpected relationships between apparently different complex phenomena, operating with different physics and at different scales 18-Apr-10 Copyright P.H.Welch 51

52 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Turbulent Flow Computer modelling and simulaton can show unexpected relationships between apparently different complex phenomena, operating with different physics and at different scales 18-Apr-10 Copyright P.H.Welch 52

53 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Turbulent Flow Computer modelling and simulaton can show unexpected relationships between apparently different complex phenomena, operating with different physics and at different scales 18-Apr-10 Copyright P.H.Welch 53

54 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Turbulent Flow Computer modelling and simulaton can show unexpected relationships between apparently different complex phenomena, operating with different physics and at different scales 18-Apr-10 Copyright P.H.Welch 54

55 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Turbulent Flow Computer modelling and simulaton can show unexpected relationships between apparently different complex phenomena, operating with different physics and at different scales 18-Apr-10 Copyright P.H.Welch 55

56 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Turbulent Flow Computer modelling and simulaton can show unexpected relationships between apparently different complex phenomena, operating with different physics and at different scales because their (differing) behavours emerge from agents following identical low-level level rules, just with slightly different key parameters 18-Apr-10 Copyright P.H.Welch 56

57 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Turbulent Flow Computer modelling and simulaton can show unexpected relationships between apparently different complex phenomena, operating with different physics and at different scales because their (differing) behavours emerge from agents following identical low-level level rules, just with slightly different key parameters 18-Apr-10 Copyright P.H.Welch 57

58 Discovering Unexpected Relations between Phenomena Starlings and Hawk Scientist / Engineer Simulation Turbulent Flow Computer modelling and simulaton can show unexpected relationships between apparently different complex phenomena, operating with different physics and at different scales because their (differing) behavours emerge from agents following identical low-level level rules, just with slightly different key parameters 18-Apr-10 Copyright P.H.Welch 58

59 Discovering and Experimenting with New Physics??? Scientist / Engineer Simulation??? Through computer modelling and simulaton,, we can investigate the emergent properties of whole new worlds of materials, new states of physics, by experimenting with varieties of agent programmed with simple low-level level rules. 18-Apr-10 Copyright P.H.Welch 59

60 Discovering and Experimenting with New Physics Free-flow Traffic Scientist / Engineer Simulation??? Through computer modelling and simulaton,, we can investigate the emergent properties of whole new worlds of materials, new states of physics, by experimenting with varieties of agent programmed with simple low-level level rules. 18-Apr-10 Copyright P.H.Welch 60

61 Discovering and Experimenting with New Physics Free-flow Traffic Scientist / Engineer Simulation Intelligent Plasma Through computer modelling and simulaton,, we can investigate the emergent properties of whole new worlds of materials, new states of physics, by experimenting with varieties of agent programmed with simple low-level level rules. Some of these may turn out to be interesting and useful so that we might be motivated to find ways to build those agents for real! 18-Apr-10 Copyright P.H.Welch 61

62 To Boldy Go Summary Research projects cosmos-research.org occam-pi.org concurrency.cc rmox.net occam-pi Kent moodle.kent.ac.uk/external/course/view.php?id=31 18-Apr-10 Copyright P.H.Welch 62

63 To Boldy Go Summary Once more, and this time with feeling 18-Apr-10 Copyright P.H.Welch 63

64 To Boldy Go Summary Future? Drug design: try to build molecules with certain shapes (to match the geometry of suspected weak spots of rogue cells) Emergent behaviours: elimination (or inhibition) of tumours. Autonomous driving: avoid collisions, head for the longest straight clear path (with speed in proportion), add bias in general favour of destination (if known) Emergent behaviours: safe driving, efficient use of the road, faster completion of journey. 18-Apr-10 Copyright P.H.Welch 64

65 Any Questions? Any Questions? Any Questions?