Wait-Free Computability for General Tasks

Slides:



Advertisements
Similar presentations
The Primal-Dual Method: Steiner Forest TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA A A AA A A A AA A A.
Advertisements

Topology in Distributed Computing: A Primer 1 / 16 Sergey Velder SPbSU ITMO.
Circuit and Communication Complexity. Karchmer – Wigderson Games Given The communication game G f : Alice getss.t. f(x)=1 Bob getss.t. f(y)=0 Goal: Find.
Tyler White MATH 493 Dr. Wanner
N-Consensus is the Second Strongest Object for N+1 Processes Eli Gafni UCLA Petr Kuznetsov Max Planck Institute for Software Systems.
Bigamy Eli Gafni UCLA GETCO Outline Models, tasks, and solvability What is SM? r/w w.f. asynch computability Sperner Lemma as a consequence of the.
Generalization and Specialization of Kernelization Daniel Lokshtanov.
Map Overlay Algorithm. Birch forest Wolves Map 1: Vegetation Map 2: Animals.
Approximations of points and polygonal chains
Complexity ©D Moshkovitz 1 Approximation Algorithms Is Close Enough Good Enough?
How should we define corner points? Under any reasonable definition, point x should be considered a corner point x What is a corner point?
CHAPTER 5: CONVEX POLYTOPES Anastasiya Yeremenko 1.
1 Discrete Structures & Algorithms Graphs and Trees: II EECE 320.
Sergio Rajsbaum 2006 Lecture 3 Introduction to Principles of Distributed Computing Sergio Rajsbaum Math Institute UNAM, Mexico.
Deriving an Algorithm for the Weak Symmetry Breaking Task Armando Castañeda Sergio Rajsbaum Universidad Nacional Autónoma de México.
1 Separator Theorems for Planar Graphs Presented by Shira Zucker.
On the Cost of Fault-Tolerant Consensus When There are no Faults Idit Keidar & Sergio Rajsbaum Appears in SIGACT News; MIT Tech. Report.
The Design and Analysis of Algorithms
Efficient Partition Trees Jiri Matousek Presented By Benny Schlesinger Omer Tavori 1.
Two-Process Systems TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A AA Companion slides for Distributed Computing.
MA5209 Algebraic Topology Wayne Lawton Department of Mathematics National University of Singapore S ,
C&O 355 Lecture 2 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A.
Manifold Protocols TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A AA Companion slides for Distributed Computing.
1 © P. Kouznetsov A Note on Set Agreement with Omission Failures Rachid Guerraoui, Petr Kouznetsov, Bastian Pochon Distributed Programming Laboratory Swiss.
C&O 355 Mathematical Programming Fall 2010 Lecture 4 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A.
Unsolvability and Infeasibility. Computability (Solvable) A problem is computable if it is possible to write a computer program to solve it. Can all problems.
C&O 355 Mathematical Programming Fall 2010 Lecture 5 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA A A.
Algorithms for hard problems WQO theory and applications to parameterized complexity Juris Viksna, 2015.
Fault tolerance and related issues in distributed computing Shmuel Zaks GSSI - Feb
FORMAL LANGUAGES, AUTOMATA AND COMPUTABILITY TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA.
1 Distributed Vertex Coloring. 2 Vertex Coloring: each vertex is assigned a color.
C&O 355 Lecture 19 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A A A A A A A A.
Topology Preserving Edge Contraction Paper By Dr. Tamal Dey et al Presented by Ramakrishnan Kazhiyur-Mannar.
Theory of Computational Complexity Yusuke FURUKAWA Iwama Ito lab M1.
Algebraic Topology Simplicial Homology Wayne Lawton
The Design and Analysis of Algorithms
On the Size of Pairing-based Non-interactive Arguments
Colorless Wait-Free Computation
Algebraic Topology and Distributed Computing part two
Byzantine-Resilient Colorless Computaton
Elements of Combinatorial Topology
Chapter 5. Optimal Matchings
Algebraic Topology and Distributed Computing
NP-Completeness Yin Tat Lee
Intro to Theory of Computation
Elements of Combinatorial Topology
Solvability of Colorless Tasks in Different Models
Distributed Consensus
Enumerating Distances Using Spanners of Bounded Degree
Distributed Consensus
Simulations and Reductions
Uri Zwick – Tel Aviv Univ.
Combinatorial Topology and Distributed Computing
Decidable Languages Costas Busch - LSU.
Algebraic Topology and Decidability in Distributed Computing
On the effect of randomness on planted 3-coloring models
Renaming and Oriented Manifolds
Combinatorial Topology and Distributed Computing
Rodolfo Conde Joint work with Sergio Rajsbaum Instituto de Matemáticas
Task Solvability in Different Models
CS21 Decidability and Tractability
Algebraic Topology and Distributed Computing part three
Instructor: Aaron Roth
NP-Completeness Yin Tat Lee
Combinatorial Topology and Distributed Computing
Combinatorial Topology and Distributed Computing
Instructor: Aaron Roth
Perfect Phylogeny Tutorial #10
Presentation transcript:

Wait-Free Computability for General Tasks Companion slides for Distributed Computing Through Combinatorial Topology Maurice Herlihy & Dmitry Kozlov & Sergio Rajsbaum TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AAAA 1

Road Map Inherently colored tasks Solvability for colored tasks Protocol ) map Map ) protocol A Sufficient Topological Conditions

Star Review ¾ Star(¾,K) is the complex of facets of K containing ¾ Distributed Computing through Combinatorial Topology

Facet Review Facet not a facet A facet is a simplex of maximal dimension Distributed Computing through Combinatorial Topology

Review Open Star Staro(¾,K) union of interiors of simplexes containing ¾ Point Set Distributed Computing through Combinatorial Topology

Review Link Link(¾,K) is the complex of simplices of Star(¾,K) not containing ¾ Complex Distributed Computing through Combinatorial Topology

A simplicial map Á is rigid if Review A simplicial map Á is rigid if dim Á(¾) = dim ¾. 18-Nov-18

Road Map Inherently colored tasks Solvability for colored tasks Protocol ) map Map ) protocol A Sufficient Topological Conditions

The Hourglass task I O

Single-Process Executions ¢ R Q R,0 Q,0 I O

P and R only (P and Q Symmetric) P P ¢ R R I O

Q and R only ¢ I O

Hourglass satisfies conditions of fundamental theorem … Claim: Hourglass satisfies conditions of fundamental theorem … But has no wait-free immediate snapshot protocol! 18-Nov-18

Hourglass satisfies conditions of fundamental theorem … Claim: Hourglass satisfies conditions of fundamental theorem … But has no wait-free immediate snapshot protocol! 18-Nov-18

homotopy: |I |  |O| O I carried by ¢

Hourglass satisfies conditions of fundamental theorem … Claim: Hourglass satisfies conditions of fundamental theorem … But has no wait-free immediate snapshot protocol! 18-Nov-18

The Hourglass task solves 2-set agreement … Claim: The Hourglass task solves 2-set agreement … Which has no wait-free read-write protocol. 18-Nov-18

Write input value to announce array … Protocol: Write input value to announce array … Run Hourglass task … 18-Nov-18

Look in announce[] array … P P P Find non-null announce[] value P,1 R or Q R Q R Q Look in announce[] array …

What Went Wrong? f: |I|  |O|... carried by ¢ Theorem A colorless (I,O,¢) has a wait-free immediate snapshot protocol iff there is a continuous map … f: |I|  |O|... We will prove a simple theorem that completely characterizes when it is possible to solve a colorless task in wait-free read-write memory. The theorem states that a colorless task $(\cI^*,\cO^*,\Delta^*)$ has a wait-free read-write protocol if and only if there is a continuous map between their polyhedrons, carried by ¢ 18-Nov-18

One Direction is OK f: |I|  |O|... carried by ¢ Theorem If (I,O,¢) has a wait-free read-write protocol … then there is a continuous map … We will prove a simple theorem that completely characterizes when it is possible to solve a colorless task in wait-free read-write memory. The theorem states that a colorless task $(\cI^*,\cO^*,\Delta^*)$ has a wait-free read-write protocol if and only if there is a continuous map between their polyhedrons, f: |I|  |O|... carried by ¢ 18-Nov-18

The Other Direction Fails Theorem? If there is a continuous map … f: |I|  |O|... carried by ¢ … We will prove a simple theorem that completely characterizes when it is possible to solve a colorless task in wait-free read-write memory. The theorem states that a colorless task $(\cI^*,\cO^*,\Delta^*)$ has a wait-free read-write protocol if and only if there is a continuous map between their polyhedrons, then does (I,O,¢) have a wait-free IS protocol? 18-Nov-18

Simplicial approximation Review f: |I|  |O|... Simplicial approximation Á: BaryN I  O Repeated snapshot Protocol 18-Nov-18

Review f: |I|  |O|... Á: BaryN I  O Simplicial approximation Not color-preserving Á: BaryN I  O Repeated snapshot Another process’s output? Protocol 18-Nov-18

Road Map Inherently colored tasks Solvability for colored tasks Protocol ) map Map ) protocol A Sufficient Topological Conditions

How can we adapt this theorem to colored tasks? A colorless (I,O,¢) has a wait-free immediate snapshot protocol iff there is a continuous map … f: |I|  |O|... carried by ¢ We will prove a simple theorem that completely characterizes when it is possible to solve a colorless task in wait-free read-write memory. The theorem states that a colorless task $(\cI^*,\cO^*,\Delta^*)$ has a wait-free read-write protocol if and only if there is a continuous map between their polyhedrons, How can we adapt this theorem to colored tasks? 18-Nov-18

Fundamental Theorem for Colored Tasks (I,O,¢) has a wait-free read-write protocol iff … I has a chromatic subdivision Div I … & color-preserving simplicial map Á: Div I  O… We will prove a simple theorem that completely characterizes when it is possible to solve a colorless task in wait-free read-write memory. The theorem states that a colorless task $(\cI^*,\cO^*,\Delta^*)$ has a wait-free read-write protocol if and only if there is a continuous map between their polyhedrons, carried by ¢ 18-Nov-18

¢ Quasi-Consensus Q,1 P,1 Q,1 P,1 P,0 Q,0 P,0 Q,0 I O Each of $P_0$ and $P_1$ is given a binary input. If both have input $v$, then both must decide $v$. If they have mixed inputs, then either they agree, or $P_1$ may decide $0$ and $P_0$ may decide 1 (but not vice-versa). P,0 Q,0 P,0 Q,0 I O

¢ Quasi-Consensus Q,1 P,1 Q,1 P,1 P,0 Q,0 P,0 Q,0 I O Each of $P_0$ and $P_1$ is given a binary input. If both have input $v$, then both must decide $v$. If they have mixed inputs, then either they agree, or $P_1$ may decide $0$ and $P_0$ may decide 1 (but not vice-versa). P,0 Q,0 P,0 Q,0 I O

¢ Quasi-Consensus P,0 Q,0 P,1 Q,1 Q,1 P,1 P,0 Q,0 I O Each of $P_0$ and $P_1$ is given a binary input. If both have input $v$, then both must decide $v$. If they have mixed inputs, then either they agree, or $P_1$ may decide $0$ and $P_0$ may decide 1 (but not vice-versa). P,0 Q,0 I O

¢ Not a colorless task! Quasi-Consensus P,0 Q,0 P,1 Q,1 Q,1 P,1 P,0

Quasi-Consensus P,0 Q,0 P,1 Q,1 Q,1 P,1 ¢ P,0 Q,0 I O

¢ No simplicial map: I  O carried by ¢ Quasi-Consensus P,0 Q,0 P,1 It is easily seen that there is no simplicial map carried by $\Delta$ from the input complex to the output complex. The vertexes of input simplex $\set{ang{P,0},\ang{Q,1}}$ $\ang{P,0}$ and $\ang{Q,1}$, but there is no single output simplex containing both vertexes. P,0 Q,0 No simplicial map: I  O carried by ¢ I O

Q,1 P,1 Q,1 P,1 Á Nevertheless, there is a map satisfying the conditions of the theorem from a subdivision of the input complex. If input simplex {<P,0>,<Q,1>} is subdivided as shown then it can be ``folded'' around the output complex, allowing input vertexes <P,0> and <Q,1> to be mapped to their counterparts in the output complex. P,0 Q,0 P,0 Q,0 Div I O

Code is asymmetric! // code for P T decide(T input) { announce[P] = input; if (input == 1) return 1; else if (announce[Q] != 1) return 0 else return 1 } Code is asymmetric! // code for Q T decide(T input) { announce[P] = input; if (input == 0) return 0; else if (announce[P] != 0) return 1 else return 0 } Here is a simple protocol for quasi-consensus. If P has input 0 and Q has input 1, then this protocol admits three distinct executions: one in which both decide 0, one in which both decide 1, and one in which Q decides 0 and P decides 1.These three executions correspond to the three simplexes in the subdivision of {<P,0>,<Q,1}>. 18-Nov-18

Road Map Inherently colored tasks Solvability for colored tasks Protocol ) map Map ) protocol A Sufficient Topological Conditions

Protocol ) Map I ¥(I) O protocol decision map ¥ δ input complex output complex carried by ¢ 18-Nov-18

Protocol ) Map ¥(I) O δ subdivision carried by of input ¢ complex 18-Nov-18

Road Map Inherently colored tasks Solvability for colored tasks Protocol ) map Map ) protocol A Sufficient Topological Conditions

¾ ¢(¾) I O Task (I,O,¢) 18-Nov-18 Recall that a task is given by an input complex \cI, an output complex \cO, and a relation \Delta that maps each input simplex \sigma to a subcomplex of \cO. ¾ ¢(¾) I O 18-Nov-18

chromatic subdivision Div ¾ chromatic simplex ¾ Recall that a that a simplex $\sigma = \set{\vs_0, \ldots, \vs_n}$ is \emph{chromatic} if each vertex is labeled with a distinct color, and a \emph{chromatic subdivision} $\Div \sigma$ is a subdivision of $\sigma$ where (1) each simplex of the subdivision is chromatic, and (2) for each $\tau \subset \sigma$, each vertex in $\Div \tau$ is labeled with a color from $\tau$. chromatic subdivision Div ¾ 18-Nov-18

The theorem we want to prove says that if there exists a chromatic subdivision Div \cI of the input complex, then there is an algorithm. ¾ ¢(¾) I O 18-Nov-18

If there is a chromatic subdivision … Theorem says … If there is a chromatic subdivision … The theorem we want to prove says that if there exists a chromatic subdivision Div \cI of the input complex, then there is an algorithm. ¾ ¢(¾) I O 18-Nov-18

à ¾ ¢(¾) I O Theorem says … If there is a chromatic subdivision … and a simplicial map à carried by ¢ … à The theorem we want to prove says that if there exists a chromatic subdivision Div \cI of the input complex, then there is an algorithm. ¾ ¢(¾) I O 18-Nov-18

à ¾ ¢(¾) I O Theorem says … If there is a chromatic subdivision … and a simplicial map à carried by ¢ … à The theorem we want to prove says that if there exists a chromatic subdivision Div \cI of the input complex, then there is an algorithm. ¾ ¢(¾) I O … then there is a wait-free IS protocol! 18-Nov-18

Let’s start with something easier … 18-Nov-18

à ChN¾ ¢(¾) I O Let’s start with a special case … If there is a simplicial map Ã: ChN ¾  ¢(¾) … à ChN¾ ¢(¾) I O 18-Nov-18 18-Nov-18 47

Á ¢(¾) ChN¾ I O Let’s start with something easier … If there is a simplicial map Á: ChN ¾  ¢(¾) … Á ¢(¾) ChN¾ I O … then there is a wait-free IS protocol! 18-Nov-18 18-Nov-18 48

Iterated immediate snapshot Protocol Iterated immediate snapshot Informally, the processes start out on vertexes of a single simplex $\sigma$ of $\cI$, and they eventually halt on vertexes of a single simplex of $\Delta(\sigma) \subseteq \cI$. 18-Nov-18

For any chromatic subdivision Div ¾ … If there is a color and carrier-preserving simplicial map Á: ChN ¾  Div ¾ … Á Div ¾ ChN¾ 18-Nov-18 18-Nov-18 50

Geometric construction Inductively divide boundary 18-Nov-18

Geometric construction Displace vetexes from barycenter 18-Nov-18

Geometric construction 18-Nov-18

Geometric construction Mesh(Ch ¾) is max diameter of a simplex 18-Nov-18

Subdivision shrinks mesh mesh(Ch ¾) · c mesh(¾) for some 0 < c < 1 18-Nov-18

Open cover An \emph{open cover} $\bbU$ for a complex $\cK$ is a finite collection of open sets $U_0, \ldots, U_k$ such that $\cK \subseteq \cup_{i=0}^k U_i$. 18-Nov-18

Lesbesgue Number ¸ 18-Nov-18 If $U_0, \ldots, U_k$ an open cover for $\cK$, there exists a $\lambda > 0$ (called a \emph{Lebesgue number}) such that any set of diameter less than $\lambda$ lies in a single $U_i$. ¸ 18-Nov-18

Open stars form an open cover for a complex ostar(v) 18-Nov-18

Vertexes lie on a common simplex iff their open stars intersect Intersection Lemma Vertexes lie on a common simplex iff their open stars intersect 18-Nov-18

… each star of ChN ¾ lies in a open star of Div ¾ Pick N large enough that each (closed) star of ChN ¾ has diameter less than ¸ … … each star of ChN ¾ lies in a open star of Div ¾ 18-Nov-18

Simplicial by intersection lemma. Defines a vertex map …. Simplicial by intersection lemma. Á 18-Nov-18

Á Div ¾ ChN¾ Á: ChN ¾  Div ¾ … We have just proved the Simplicial Approximation Theorem Á Div ¾ ChN¾ There is a carrier-preserving simplicial map Á: ChN ¾  Div ¾ … Not necessarily color-preserving! 18-Nov-18 18-Nov-18 62

An open-star cover is chromatic if every simplex ¿ of ChN ¾ is covered by open stars of of the same color. 18-Nov-18

If the open-star cover is chromatic …. Then the simplicial map …. Is color preserving! Á Must show that covering can be made chromatic … 18-Nov-18

Open Cover Fail Two simplexes conflict … If colors disjoint, but … polyhedrons overlap. cannot map to same color 18-Nov-18

We will show how to eliminate conflicting simplexes Open Cover Fail An open-star cover is chromatic iff there are no conflicting simplexes. We will show how to eliminate conflicting simplexes 18-Nov-18

Carriers Let $\Div \sigma$ be a subdivision of $\sigma$. For any simplex $\tau \in \Div \sigma$, the \emph{carrier} of $\tau$ in $\Div \sigma$ is the smallest face $\kappa$ of $\sigma$ such that $\tau$ is in $\Div \kappa$. 18-Nov-18

Perturbation 18-Nov-18

Star contains ² ball in carrier around vertex Room for perturbation Star contains ² ball in carrier around vertex 18-Nov-18

Room for perturbation Can perturb to any point within ² ball in carrier and still have subdivision 18-Nov-18

Open Cover Fail ½ has q+1 colors ¿ has p+1 colors 18-Nov-18

Simplexes lie in hyperplane of dimension p+q (because they overlap) 18-Nov-18

Some vertex has carrier of dimension p+q+1 (because there are p+q+2 colors) 18-Nov-18

Can perturb vertex within (p+q+1)-dimension ² ball … 18-Nov-18

Can perturb vertex within (p+q+1)-dimension ² ball … Out of the hyperplane 18-Nov-18

Repeat until star diameter < Lebesgue number: Construct Ch ChN-1* ¾ Perturb to ChN* ¾ So open-star cover is chromatic Construct color-preserving simplicial map 18-Nov-18

Div ¾ ¢(¾) Ã Given 18-Nov-18 18-Nov-18 77

ChN¾ Constructed Á Div ¾ ¢(¾) Ã Given 18-Nov-18 18-Nov-18 78

Iterated immediate snapshot here … ChN¾ Á Div ¾ ¢(¾) Ã Yields protocol here! 18-Nov-18 18-Nov-18 79

Road Map Inherently colored tasks Solvability for colored tasks Protocol ) map Map ) protocol A Sufficient Topological Condition

Link-Connected O is link-connected if for each ¿ 2 O, A geometric complex is subdivided by partitioning each of its simplexes into smaller simplexes without changing the complex's polyhedron. O is link-connected if for each ¿ 2 O, link(¿,O) is (n - 2 - dim ¿)-connected. 18-Nov-18

not a fan not link-connected. 18-Nov-18

Theorem If, for all ¾ 2 I, ¢(¾) is ((dim ¾)-1)-connected, and O is link-connected then (I,O,¢) has a wait-free IS protocol 18-Nov-18

Proof Strategy If, for all ¾ 2 I, ¢(¾) is ((dim ¾)-1)-connected, and O is link-connected, there exists subdivision Div & color-preserving simplicial map ¹: Div I ! O carried by ¢. 18-Nov-18

Proof Strategy If, for all ¾ 2 I, ¢(¾) is ((dim ¾)-1)-connected, and so protocol exists by colored theorem O is link-connected, there exists subdivision Div & color-preserving simplicial map ¹: Div I ! O carried by ¢. 18-Nov-18

Lemma rigid & color-preserving on boundary means color-preserving everywhere suppose we have a rigid simplicial map Á: Div ¾  O that is color-preserving on Div  ¾ then Á is color-preserving on Div ¾ 18-Nov-18

Lemma If O is link-connected … can extend rigid simplicial map Án-1: skeln-1 I  O to a rigid simplicial map Án: Div I  O where Div skeln-1 I = skeln-1 I 18-Nov-18

Á O I O Á’ div I Induction Base: n = 1 collapses hinge does not collapse0.eps Á’ div I does not collapse

Induction Step: Á does not collapse (n-1)-simplexes

Div I collapse2.eps O

Div I does not collapse Á’ O exploit connected link

Summary connectivity of ¢(¾) Inductively use … link-connectivity of O to construct a color-preserving simplicial map Á: Div I  Ocarried by ¢. protocol follows from main theorem 18-Nov-18

          This work is licensed under a Creative Commons Attribution-Noncommercial 3.0 Unported License.