EE384y: Packet Switch Architectures Matchings, implementation and heuristics High Performance Switching and Routing Telecom Center Workshop: Sept 4, 1997. Nick McKeown Professor of Electrical Engineering and Computer Science, Stanford University nickm@stanford.edu www.stanford.edu/~nickm Spring 2008 1 Outline Finding a maximum match. Maximum network flow problems Maximum size/weight matchings as examples of maximum network flows • Definitions and example • Augmenting paths • Maximum size matching • Complexity of maximum size matchings and maximum weight matchings What algorithms are used in practice? Maximal Matches Wavefront Arbiter (WFA) Parallel Iterative Matching (PIM) iSLIP Spring 2008 2 Network Flows Source s 10 10 • • • a c 10 1 b 1 1 d 10 10 Sink t Let G = [V,E] be a directed graph with capacity cap(v,w) on edge [v,w]. A flow is an (integer) function, f, that is chosen for each edge so that f (v, w) cap(v, w). We wish to maximize the flow allocation. Spring 2008 3 A maximum network flow example By inspection Source s a 10 10 1 b Step 1: Source s a 1 1 10, 10 b d c 1 10, 10 10 c 10 1 1 d 10 10 10, 10 10 Sink t Sink t Flow is of size 10 Spring 2008 4 A maximum network flow example Step 2: Source s a c 1 10, 10 10, 1 10, 10 b 1 1, 1 d 10, 10 Sink t 10, 1 Flow is of size 10+1 = 11 Maximum flow: Source s a 10, 10 10, 2 b 10, 9 c 1,1 1,1 1, 1 d 10, 10 Not obvious Sink t 10, 2 Flow is of size 10+2 = 12 Spring 2008 5 Ford-Fulkerson method of augmenting paths 1. 2. 3. 4. 5. Set f(v,w) = -f(w,v) on all edges. Define a Residual Graph, R, in which res(v,w) = cap(v,w) – f(v,w) Find paths from s to t for which there is positive residue. Increase the flow along the paths to augment them by the minimum residue along the path. Keep augmenting paths until there are no more to augment. Spring 2008 6 Example of Residual Graph a s 10, 10 1 10, 10 10 b c 1 1 d 10, 10 10 t Flow is of size 10 Residual Graph, R res(v,w) = cap(v,w) – f(v,w) a 10 c 10 s 10 10 1 b 1 1 d 10 t Augmenting path Spring 2008 7 Example of Residual Graph Step 2: a s c 1 10, 10 10, 1 10, 10 b 1 d 1, 1 10, 10 t 10, 1 Flow is of size 10+1 = 11 Residual Graph 10 s 1 9 a c 10 1 b 1 1 d 10 t 1 9 Spring 2008 8 Example of Residual Graph Step 3: a s 1, 1 10, 10 10, 2 c 10, 9 b 1, 1 1, 1 d 10, 10 t 10, 2 Flow is of size 10+2 = 12 Residual Graph 10 s 2 8 a 1 c 9 1 b 1 1 d 10 t 2 8 Spring 2008 9 Complexity of network flow problems In general, it is possible to find a solution by considering at most |V|.|E| paths, by picking shortest augmenting path first. There are many variations, such as picking most augmenting path first. Spring 2008 10 Outline Finding a maximum match. Maximum network flow problems Maximum size/weight matchings as examples of maximum network flows • Definitions and example • Augmenting paths • Maximum size matching • Complexity of maximum size matchings and maximum weight matchings What algorithms are used in practice? Maximal Matches Wavefront Arbiter (WFA) Parallel Iterative Matching (PIM) iSLIP Spring 2008 11 Finding a maximum size match A 1 B 2 C 3 D 4 E 5 F 6 How do we find the maximum size (weight) match? Spring 2008 12 Network flows and bipartite matching Source s A 1 B 2 C 3 D 4 E 5 F 6 Sink t Finding a maximum size bipartite matching is equivalent to solving a network flow problem with capacities and flows of size 1. Spring 2008 13 Network flows and bipartite matching Ford-Fulkerson method Residual Graph for first three paths: s A 1 B 2 C 3 D 4 E 5 F 6 t Spring 2008 14 Network flows and bipartite matching Residual Graph for next two paths: s A 1 B 2 C 3 D 4 E 5 F 6 t Spring 2008 15 Network flows and bipartite matching Residual Graph for augmenting path: s A 1 B 2 C 3 D 4 E 5 F 6 t Spring 2008 16 Network flows and bipartite matching Residual Graph for last augmenting path: s A 1 B 2 C 3 D 4 E 5 F 6 t Note that the path augments the match: no input and output is removed from the match during the augmenting step. Spring 2008 17 Network flows and bipartite matching Maximum flow graph: s A 1 B 2 C 3 D 4 E 5 F 6 t Spring 2008 18 Network flows and bipartite matching Maximum Size Matching: A 1 B 2 C 3 D 4 E 5 F 6 Spring 2008 19 Complexity of Maximum Matchings Maximum Size Matchings: Maximum Weight Matchings Algorithm by Dinic O(N5/2) Algorithm by Kuhn O(N3) In general: Hard to implement in hardware Slooooow. Spring 2008 20 Outline Finding a maximum match. Maximum network flow problems Maximum size/weight matchings as examples of maximum network flows • Definitions and example • Augmenting paths • Maximum size matching • Complexity of maximum size matchings and maximum weight matchings What algorithms are used in practice? Maximal Matches Wavefront Arbiter (WFA) Parallel Iterative Matching (PIM) iSLIP Spring 2008 21 Maximal Matching A maximal matching is one in which each edge is added one at a time, and is not later removed from the matching. i.e. no augmenting paths allowed (they remove edges added earlier). No input and output are left unnecessarily idle. Spring 2008 22 Example of Maximal Size Matching A 1 A 1 A 1 B 2 B 2 B 2 C 3 C 3 C 3 D 4 D 4 4 E 5 D E 5 5 F 6 F 6 E F 6 Maximal Size Matching Maximum Size Matching Spring 2008 23 Maximal Matchings In general, maximal matching is simpler to implement, and has a faster running time. A maximal size matching is at least half the size of a maximum size matching. A maximal weight matching is defined in the obvious way. A maximal weight matching is at least half the weight of a maximum weight matching. Spring 2008 24 Outline Finding a maximum match. Maximum network flow problems Maximum size/weight matchings as examples of maximum network flows • Definitions and example • Augmenting paths • Maximum size matching • Complexity of maximum size matchings and maximum weight matchings What algorithms are used in practice? Maximal Matches Wavefront Arbiter (WFA) Parallel Iterative Matching (PIM) iSLIP Spring 2008 25 Wave Front Arbiter (Tamir) Requests Match 1 1 1 1 2 2 2 2 3 3 3 3 4 4 4 4 Spring 2008 26 Wave Front Arbiter Requests Match Spring 2008 27 Wave Front Arbiter Implementation 1,1 1,2 1,3 1,4 2,1 2,2 2,3 2,4 3,1 3,2 3,3 3,4 4,1 4,2 4,3 4,4 Simple combinational logic blocks Spring 2008 28 Wave Front Arbiter Wrapped WFA (WWFA) N steps instead of 2N-1 Requests Match Spring 2008 29 Wavefront Arbiters Properties Feed-forward (i.e. non-iterative) design lends itself to pipelining. Always finds maximal match. Usually requires mechanism to prevent Q11 from getting preferential service. In principle, can be distributed over multiple chips. Spring 2008 30 Outline Finding a maximum match. Maximum network flow problems Maximum size/weight matchings as examples of maximum network flows • Definitions and example • Augmenting paths • Maximum size matching • Complexity of maximum size matchings and maximum weight matchings What algorithms are used in practice? Maximal Matches Wavefront Arbiter (WFA) Parallel Iterative Matching (PIM) iSLIP Spring 2008 31 Parallel Iterative Matching uar selection uar selection #1 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 4 4 4 4 4 4 f1: Requests f2: Grant f3: Accept/Match 1 2 #2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 4 4 4 4 4 4 Spring 2008 32 PIM Properties Guaranteed to find a maximal match in at most N iterations. In each phase, each input and output arbiter can make decisions independently. In general, will converge to a maximal match in < N iterations. How many iterations should we run? Spring 2008 33 Parallel Iterative Matching Convergence Time Number of iterations to converge: N2 E U i ------4i E C log N C = # of iterations required to resolve connections N = # of ports U i = # of unresolved connections after iteration i Spring 2008 34 Parallel Iterative Matching Spring 2008 35 Parallel Iterative Matching PIM with a single iteration Spring 2008 36 Parallel Iterative Matching PIM with 4 iterations Spring 2008 37 Outline Finding a maximum match. Maximum network flow problems Maximum size/weight matchings as examples of maximum network flows • Definitions and example • Augmenting paths • Maximum size matching • Complexity of maximum size matchings and maximum weight matchings What algorithms are used in practice? Maximal Matches Wavefront Arbiter (WFA) Parallel Iterative Matching (PIM) iSLIP Spring 2008 38 iSLIP 1 4 #1 1 2 3 1 2 3 1 2 3 1 2 3 4 4 4 4 F1: Requests F2: Grant 2 3 1 2 3 1 2 3 4 4 F3: Accept/Match 1 2 #2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 4 4 4 4 4 4 Spring 2008 39 iSLIP Operation Grant phase: Each output selects the requesting input at the pointer, or the next input in round-robin order. It only updates its pointer if the grant is accepted. Accept phase: Each input selects the granting output at the pointer, or the next output in round-robin order. Consequence: Under high load, grant pointers tend to move to unique values. Spring 2008 40 iSLIP Properties Random under low load TDM under high load Lowest priority to MRU 1 iteration: fair to outputs Converges in at most N iterations. (On average, simulations suggest < log2N) Implementation: N priority encoders 100% throughput for uniform i.i.d. traffic. But…some pathological patterns can lead to low throughput. Spring 2008 41 iSLIP Spring 2008 42 iSLIP Spring 2008 43 Programmable Priority Encoder N N iSLIP Implementation 1 1 Grant Accept 2 2 Grant Accept log2N log2N State Decision N N Grant N Accept log2N Spring 2008 44 Maximal Matches Maximal matching algorithms are widely used in industry (PIM, iSLIP, WFA and others). PIM and iSLIP are rarely run to completion (i.e. they are sub-maximal). A maximal match with a speedup of 2 is stable for non-uniform traffic. Spring 2008 45