More is Different

What on earth are scale and complexity!

Transcript

1. More Is Different Nigel Warren
2. Topics ▪ Philip Anderson - More is Different ▪ Emergence and Complexity ▪ Cellular Automata ▪ Turing, von Neumann and Beyond ▪ Fallacies of Distributed Computing ▪ Boundaries - Laws and Proofs ▪ Biological Analogy ▪ Scale and Complexity ▪ Round Up
3. Scale and Complexity ‘The ability to reduce everything to simple fundamental laws does not by any means imply the ability to start from those laws and reconstruct the universe.’ ‘The Constructionist hypothesis breaks down when confronted with the twin difficulties of scale and complexity’ More Is Different - P. W. Anderson Science, New Series, Vol. 177, No. 4047. (Aug. 4, 1972), pp. 393-396.
4. Cellular Automata Established by Stanislaw Ulam and John von Neumann Los Alamos National Lab ‘Could a machine build copies of itself and succeed’ Simplified by Conway’s Game of Life John Conway - John von Neumann Chair of Mathematics at Princeton University Simple and a Universal Turing Machine. I.e. Anything that can be computed can be computed in the game of life
5. Cellular Automata
6. The Most Simple Automata Elementary Cellular Automaton Binary Nearest Neighbour One Dimensional Rule 30 for example, of 256 possible rules
7. The Most Simple Automata Rule 30
8. Outcomes Some rules result in non reversible random structures Some rules result in complex non repeating structures Rule 110 has been proven to be Universal I.e. Anything that can be computed can be computed by Rule 110
9. Single Thread ▪ What could possibly go wrong ?
10. Many Threads
11. Many Threads - Many Cores
12. Distributed System A B
13. Distributed System A Network B
14. Fallacies of Distributed Systems L. Peter Deutsch Bill Joy James Gosling
15. The Network is Reliable
16. Latency is Zero
17. Bandwidth is Infinite
18. The Network is Secure
19. Topology Doesn’t Change
20. There is One Administrator
21. Transport Cost is Zero
22. The Network is Homogeneous
23. Its worse than that! These problems compound One problem can cause another More than one problem may occur
24. Any Boundaries ? FLP Consensus Impossibility - Fisher, Lynch & Paterson CAP Theorem - Eric Brewer
25. FLP Consensus Proof Michael J. Fischer, Nancy Lynch, and Mike Paterson It is assumed that nodes can only fail by crashing; that the network is reliable, and that the typical timing assumptions of the asynchronous system model hold, i.e. there are no bounds on message delay then … ‘No algorithm can always reach consensus in bounded time’
26. C.A.P. Consistency – All nodes share the same state Availability – Every Request receives a response Partition Tolerance – Partial failure
27. Consistency Availability Partition Tolerance
28. Failure N = 1000 MTTF = 365 * 3 =~ 1000 failure rate = 1 per day
29. Systems Lifetimes The whole system must out-live the failure of some/any single component.
30. Biological Analogy Allan Kay - The Computer Revolution hasn't happened yet.
31. Scale and Complexity
32. Thank you
33. References Distributed Systems for Fun and Profit - by Mikito Takada A Note on Distributed System – Jim Waldo et al CAP Twelve Years Later : How the "Rules" Have Changed – Eric Brewer Allan Kay - The Computer Revolution hasn't happened yet. Signals and Boundaries – John H Holland Anti Fragile – Nassim Nicholas Taleb What is Life – Erwin Schrodinger