# © Sudhakar Yalamanchili, Georgia Institute of Technology (except as indicated) Routing Algorithms- I.

## Presentation on theme: "© Sudhakar Yalamanchili, Georgia Institute of Technology (except as indicated) Routing Algorithms- I."— Presentation transcript:

© Sudhakar Yalamanchili, Georgia Institute of Technology (except as indicated) Routing Algorithms- I

ECE 8813a (2) Reading Assignment Sections 4.1-4.5 Alternative: relevant papers.  References provided

ECE 8813a (3) Overview Separate the routing decisions from implementation  Choice of path vs.  Performance oE.g., buffering, route computation implementation Routing Algorithm fixed by  Routing function  Selection function Primary correctness concern  Deadlock and livelock freedom Focus in key principles  Not coverage of all algorithms designed to date A single function for deterministic protocols

ECE 8813a (4) Virtual Channel Router S D Note: Separation of physical and virtual paths, e.g., shortest path may traverse multiple virtual networks

ECE 8813a (5) Taxonomy Routing Algorithms Unicast multicast CentralizedSourceDistributedMulti-phase Table LookupFinite State Machine DeterministicAdaptive ProgressiveBacktracking ProfitableMisrouting CompletePartial # Destinations Routing Decisions Implementation Adaptivity Progressiveness Minimality Number of Paths

ECE 8813a (6) Classes of Routing Algorithms Deterministic and Oblivious Routing  Deterministic protocols are oblivious oConverse may not be true  Variety of implementations oSource Routing, finite state machines, interval routing, table-lookup Partially Adaptive Fully, maximal, and true fully adaptive  Fully: maximize alternative physical paths  Maximal: maximize all routing options  True: no constraints on VC usage Non-minimal Routing Hybrid or multi-phase protocols

ECE 8813a (7) Deterministic Routing Algorithms Strictly increasing or decreasing order of dimension Routing function always returns the same output channel for each destination Mesh Binary Hypercube Deterministic routing

ECE 8813a (8) Tori Dimension order routing Create an acyclic channel dependency graph with the following routing function - c 0i when j i n0n1 n3n2 c0c0 c1c1 c2c2 c3c3 c 13 n0n1 n3n2 c 10 c 11 c 02 c 12 c 01 c 03 c 00 c 10 c 11 c 12 c 01 c 02 c 03 Deterministic routing

ECE 8813a (9) Physical Router Finite state machine or Table- look up Deterministic routing Distributed Routing

ECE 8813a (10) Virtual Channel Router Finite state machine or Table-look up Performance constraints Remember, we can have virtual lanes Deterministic routing Distributed Routing

ECE 8813a (11) Deterministic Routing Algorithms: Implementation Issues Relatively, the most inexpensive to implement  Each node implements a routing function oShared or private logic across channels  Absolute vs. relative addressing Header update  Necessary for relative addressing  Necessary to maintain uniformity of implementation Implementation  Table look-up vs. finite state machine Deterministic routing

ECE 8813a (12) Source Routing All routes are pre-computed at the source  Stored as a sequence of port addresses at intermediate routers Each intermediate router uses the header flit to identify the output port  Can be extended to use virtual channels ENW index output port Oblivious routing

ECE 8813a (13) Characteristics Simple fast, route computation at intermediate routers  Header update Oblivious routing  The source-destination pair does not need to identify a unique path  not deterministic Complexity of computing paths  Ensuring deadlock freedom? Extensions to virtual channels Oblivious routing We will come back to this

ECE 8813a (14) Filling in the Routing Tables How do you find deadlock free paths?  Non-trivial problem Short-cuts  Obtain a channel dependency graph for a specific algorithm  Find paths in channel dependency graphs!  End with virtual channel assignment N4 E1 S0 S1 index index at next node Oblivious routing From M. Kinsy, et.al., ”Application Aware, Deadlock Free Oblivious Routing,” Proceedings of ISCA 2009

ECE 8813a (15) Application-Specific Oblivious Routing Channel Dependence Graph Flow Graph Two different Channel Dependency Graphs From M. Kinsy, et.al., ”Application Aware, Deadlock Free Oblivious Routing,” Proceedings of ISCA 2009 Oblivious routing

ECE 8813a (16) Interval Routing Each output link corresponds to an interval of nodes  Union of intervals at a node is the set of all destination nodes  Must be able to distinguish invalid intervals (at the edges of a mesh) Use overlapping intervals for fault tolerance 04812 15913 261014 371115 ChannelNodeInterval X+68-15 X-60-3 Y+64-5 Y-67 Oblivious routing J.V. Leeuwen and R. B. Tan, “Interval Routing,” The Computer Journal, vol. 30, no.4, 1987

ECE 8813a (17) Partially Adaptive Routing Algorithms Trade-off between hardware resources and adaptivity  Maximize adaptivity for given resources  Minimize resources required for a given level of adaptivity Typically exploit regular topologies Partially Adaptive Routing A. A. Chien and J. H. Kim, “Planar Adaptive Networks: Low Cost Adaptive Networks for Multiprocessors, ISCA 1992

ECE 8813a (18) Planar Adaptive Routing Packets are routed adaptively in a series of two dimensional planes  Order of planes (dimensions) is arbitrary Routing in two dimension uses two virtual networks  Increasing and decreasing networks Fully adaptive Planar adaptive A0 A1 A0 A1 A2 Partially Adaptive Routing

ECE 8813a (19) Adaptive Routing in Two Dimensions 0123 4567 891011 12131415 0123 4567 891011 12131415 Increasing Network Decreasing Network D i+1 DiDi Partially Adaptive Routing

ECE 8813a (20) PAR in Multidimensional Networks Routing is fully adaptive in a plane When can you skip a plane? Dimension i Dimension i+1 Dimension i+2 A1 A0 Partially Adaptive Routing

ECE 8813a (21) PAR Properties Each plane is comprised of the following channels A i = d i,2 + d i+1,0 + d i+1,1 Three virtual channels/link in meshes and six virtual channels/link in Tori Partially Adaptive Routing

ECE 8813a (22) The Turn Model What is a turn?  From one dimension to another : 90 degree turn  To another virtual channel in the same direction: 0 degree turn  To the reverse direction: 180 degree turn Turns combine to form cycles Goal: prohibit the least number of turns to break all possible cycles abstract cyclesXY routingwest-first routing Partially Adaptive Routing C. J. Glass and L. Ni, “The Turn Model for Adaptive Routing,” ISCA 1992

ECE 8813a (23) Turn Constraints Choice of prohibited turns is not arbitrary Alternative designs  Three combinations unique (within symmetry) oThree algorithms: west-first, north-last, negative-first equivalent Partially Adaptive Routing

ECE 8813a (24) West First Routing Fully adaptive to the east, deterministic to the west (for non-minimal routing) Non-minimal is partially adaptive to the west deterministic region fully adaptive region Partially Adaptive Routing

ECE 8813a (25) Generalization of the Turn Model Identify channel classes and prohibit turns between them Cycles are infeasible within a channel class Transitions between channel classes are acyclic Partially adaptive: number of shortest paths are reduced Application to Binary hypercubes Base is e-cube routing Identify up channels and down channels Adaptively route though one class and then the other Turns prohibited from one class to the other Partially Adaptive Routing

ECE 8813a (26) P-Cube Routing For a given source destination pair compute the up channel set and down channel set Restrict turns from one set to the other Adaptively route within a set The escape path corresponds to some fixed order which remains a subset of channels 0000 0010 1111 1101 up channel down channel

ECE 8813a (27) Fully Adaptive Routing Using all available paths  Minimal vs. non-minimal Distinguish between buffer-based schemes vs. channel-based schemes  Typically the former can be translated to the latter Fully Adaptive Routing

ECE 8813a (28) Structured Buffer Pools Positive hop algorithm  D+1 buffers at each node (D = diameter)  Nodes request/use buffers in strictly increasing order  Minimal path algorithm, valid for any topology  Large buffer requirements: O(Diameter) buffer 0 buffer 1 buffer D-1 Fully Adaptive Routing I. S. Gopal, “Prevention of Store and Forward Deadlock in Computer Networks,” IEEE Transactions on Communication, December 1985.

ECE 8813a (29) Structured Buffer Pools: Extension subset S-1subset 1subset 0 Negative hop algorithm  Partition nodes into non-adjacent subsets  Order subsets  Down transitions request higher numbered buffer, else same numbered buffer  Number of buffers required in each node is given by + 1 set 0 set 1 Fully Adaptive Routing

ECE 8813a (30) Extensions to Wormhole Switching replace central buffers with equivalent number of virtual channels across each physical channel Acyclic central buffer dependencies are transformed into acyclic virtual channel dependencies Basic version produces unbalanced use of virtual channels  distance rarely equal to diameter Extension: bonus cards.  Number is equal to unused hops: diameter - #req  Use this number to increase the number of choices of virtual channels at any node  Not the same as adaptivity Fully Adaptive Routing

ECE 8813a (31) Virtual Networks Establish multiple virtual networks  Each network works for a specific destination set  Routing functions in a network are acyclic, but are typically not connected Establish routing constraints between virtual networks Simple customized protocols  Expensive in terms of virtual channels Fully Adaptive Routing

ECE 8813a (32) Virtual Networks in a Mesh 0123 4567 891011 12131415 0123 4567 891011 12131415 Y X X+Y+X-Y+ Each virtual network is constructed to have acyclic channel dependencies  Routing function in a virtual network is not connected Packets are injected into the appropriate virtual network Fully adaptive, no transitions between networks 2 n virtual networks with (n.2 n ) virtual channels/node Fully Adaptive Routing

ECE 8813a (33) References C.R. Jesshope, P.R. Miller, and J.T. Yantchev, “High performance communications in processor networks,'‘ Proceedings of the 16th International Symposium on Computer Architecture, pp.150-157, May-June 1989 D.H. Linder and J.C. Harden, ``An adaptive and fault tolerant wormhole routing strategy for k-ary n-cubes,'‘ IEEE Transactions on Computers, vol. C-40, no. 1, pp.2-12, January 1991

ECE 8813a (34) Virtual Networks in a Mesh 0123 4567 891011 12131415 0123 4567 891011 12131415 Y X X+Y+X-Y+ Extensions to 3D Extensions to tori and irregular topologies Couple with source routing? Extensions to sub-topologies Limit adaptivity to a subset of nodes

ECE 8813a (35) Optimize Reduce the number of networks by 50% with one additional channel in dimension 0 Total number of channels/node  Number of channels in each network + extra channel in dimensions 0 (in networks 0123 4567 891011 12131415 0123 4567 891011 12131415 Virtual network 0Virtual network 1 Fully Adaptive Routing

ECE 8813a (36) Optimize Extensions to k-ary n-cubes by introducing levels in each virtual network  One level for each wrap-around channel  At most n levels are traversed (in addition to the original level)  (n+1).2 (n-1) virtual channels/physical channel for dimensions >0 0123 4567 891011 12131415 0123 4567 891011 12131415 Virtual network 0Virtual network 1 Fully Adaptive Routing

ECE 8813a (37) Extensions to Tori Addition of layers (virtual networks) for each wrap- around connection Number of layers increases by #dimensions Fully Adaptive Routing

ECE 8813a (38) Using Dynamic Message Dependencies Two virtual channel classes across each physical link  Adaptive channels & deterministic channels Fully adaptive use of adaptive channels  Keep track of #dimension reversals for each message oMoving from a dimension p to a lower dimension q  Label each channel with the DR# of the message  Messages cannot block on a channel with lower DR# oIf no channel available, permanent transition to the deterministic channel  Dependencies between messages are acyclic Fully Adaptive Routing W.J. Dally and H. Aoki, “Deadlock-free adaptive routing in multicomputer networks using virtual channels,” IEEE Transactions on Parallel and Distributed Systems, vol. 4, no. 4, pp. 466-475, April 1993.

ECE 8813a (39) Summary of Design Techniques Ordered use of topological features  Dimensions and paths Ordered use of resources  Buffers/channels, and networks Order the message population  Each message is uniquely identified by some attribute, e.g., number of wrap-around channel crossings  Order blocking based on message population membership Design

ECE 8813a (40) Design Methodology Start with a network, set of channels C 1, and routing function R 1  R 1 is connected and deadlock free and may be deterministic/adaptive, minimal/non-minimal Split each physical channel into a set of additional virtual channels and define the new routing function  Set of channels includes escape channels and adaptive channels  Selection function can be defined in many ways For wormhole switching, verify that the extended channel dependency graph is acyclic: likely if R is restricted to minimal paths Design Set of channels from node x (local node) to node y)

ECE 8813a (42) Maximally Adaptive Routing Establish a relationship between routing freedom and resources  Maximize adaptivity for fix resources  Minimize resources for target adaptivity Relationship between adaptivity and performance  Not obvious  Unbalanced use of physical or virtual channel resources Maximally Adaptive Routing

ECE 8813a (43) References C.J. Glass and L.M. Ni, “Maximally fully adaptive routing in 2D meshes,” Proceedings of the International Conference on Parallel Processing, August 1992. R. Cypher and L. Gravano, “Storage-efficient, deadlock-free packet routing algorithms for torus networks,” IEEE Transactions on Computers, vol. 43, no. 12, pp.1376-1385, December 1994 L. Schwiebert and D.N. Jayasimha, “Optimal fully adaptive wormSupercomputing'93hole routing for meshes,”, pp.782-791, November 1993

ECE 8813a (44) In 2D Meshes: Double Y Maximally Adaptive Routing 0123 4567 891011 12131415 0123 4567 891011 12131415 increasing networkdecreasing network X+ X-Y1- Y1+ Y1- X- X+ Y2+ Y2- X-X+ X- Y2+ X+Y2- Y1Y2 6 Y1

ECE 8813a (45) In 2D Meshes: Mad-Y X+ X-Y1- Y1+ Y1- X- X+ Y2+ Y2- X-X+ X- Y2+ X+Y2- Permit turns  From the Y1 channels to the X+ channels  From X- channels to Y2 channels Remove unnecessary turn restrictions Still overly restrictive! permitted 6 Y2 Y1 No coupling between Y2 and Y1 Maximally Adaptive Routing

ECE 8813a (46) In 2D Meshes: Opt Y X+ X-Y1- Y1+ Y1- X- X+ Y2+ Y2- X-X+ X- Y2+ X+Y2- Y2+Y1+ Y2+ Y2-Y1- Y2- Further reduce the number of restrictions  Only restrict turns from Y1 to X-  Turns from X- to Y1 and 0-degree turns in Y only when X offset is 0 or positive Extensions to multidimensional meshes Basic idea: fully adaptive routing in one set of channels, and dimension order in the other set until specified lower dimension traversals are complete Maximally Adaptive Routing

ECE 8813a (47) Routing with Minimum Buffer Requirements Key Idea:  Organize packet traffic into disjoint groups that use separate buffers in each node  Place acyclic routing restrictions in buffer usage Based on node orderings Maximally Adaptive Routing

ECE 8813a (48) Node Labeling for 2D Torus 0123 4567 891011 12131415 141312 111098 7654 3210 15141213 111089 3201 7645 0132 4576 12131514 891110 right increasing node ordering left increasing node ordering outside increasing node orderinginside increasing node ordering Maximally Adaptive Routing

ECE 8813a (49) Algorithm 1 Algorithm  Packet moves from the injection queue to the A queue can  Stay in the A queues as long as we can move to right along at least one dimension along a minimal path  Transition to the B queues under same rule for left traversals  Transition to the C queue and remain there until packet is delivered Note the de-coupling of node labeling from buffer labeling Maximally Adaptive Routing

ECE 8813a (50) Application 0 1 2 4 5 6 7 3 76543210 Orderings analogous to virtual planes  Note the orderings are acyclic Extensions to edge buffers  Check Algorithm 2 Maximally Adaptive Routing