Irregular network

Abstract

Irregularities are provided in at least one dimension of a torus or mesh network for lower average path length and lower maximum channel load while increasing tolerance for omitted end-around connections. In preferred embodiments, all nodes supported on each backplane are connected in a single cycle which includes nodes on opposite sides of lower dimension tori. The cycles in adjacent backplanes hop different numbers of nodes.

Description

BACKGROUND OF THE INVENTION

An interconnection network consists of a set of nodes connected by channels. Such networks are used to transport data packets between nodes. They are used, for example, in multicomputers, multiprocessors, and network switches and routers. In multicomputers, they carry messages between processing nodes. In multiprocessors, they carry memory requests from processing nodes to memory nodes and responses in the reverse direction. In network switches and routers, they carry packets from input line cards to output line cards. For example, pending U.S. patent application Ser. No. 08/918,556, filed by William J. Dally, Philip P. Carvey, Larry R. Dennison and P. Alan King for Internet Switch Router, describes the use of a three-dimensional torus interconnection network to provide the switching fabric for an internet router.

The arrangement of the nodes and channels in a network defines its topology. FIG. 1 , for example, illustrates a two-dimensional, radix-8 torus topology, also called an 8-ary 2-cube. In this topology each node is designated by a two-digit radix-8 address. Channels in both directions connect each pair of nodes whose addresses differ by one in exactly one digit modulo 8. That is, node (a,b) is connected to nodes (a−1,b),(a+1,b),(a,b−1), and (a,b+1) where the addition and subtraction are performed modulo-8. Each node in this topology has degree 4 since it is connected to four other nodes. A related topology, an 8-ary 2-mesh, is shown in FIG. 2 . This network is identical to the 8-ary 2-cube except that the ‘end-around’ connections from node 7 to node 0 in each dimension are omitted. While the mesh network is simpler to wire, it has significantly lower performance than the torus network as will be described below.

Meshes and tori may be constructed with different radices and numbers of dimensions, and other network topologies are possible such as crossbars, trees, and butterfly networks. Several popular topologies are described, for example, by William J. Dally, “Network and Processor Architectures for Message-Driven Computers,” in VLSI and Parallel Computation , Edited by Suaya and Birtwistle, Morgan Kaufmann, 1990.

The topology chosen for an interconnection network is constrained by the technology used to implement the network. Pin-count constraints on chips, circuit boards, and backplanes, for example, limit the maximum degree of each network node. Wire-length constraints limit the maximum distance a single channel can traverse when the network is mapped into a physical packaging structure.

Low-dimensional (2-3 dimensions) torus and mesh networks have become popular in recent years in part because they offer good performance while having a small node degree and uniformly short wires when mapped to two or three dimensional electronic packaging systems. While at first it may appear that the end-around channels in a torus network result in long wires, these long wires can be avoided by folding the torus as illustrated in FIG. 3 . The figure shows how an 8×4 torus that is folded in the x-dimension is mapped onto a set of backplanes 20 . Each node is packaged on a circuit card and eight of these circuit cards are inserted in each backplane 20 . The circuit cards in each backplane span two x coordinates and the entire y dimension. The y-dimension channels are realized entirely in the backplane while the x-dimension channels are realized by connections between backplanes. Folding the torus allows this topology to be realized with each x-channel connecting no further than the adjacent backplane.

To expand the network of FIG. 3 , it is necessary to first break the loopback connection 22 or 24 at the end of the x-dimension and then make a connection to a new backplane. An efficient method for reconfiguring the network in this manner is disclosed in pending U.S. patent application Ser. No. 09/083,722, filed by Philip P. Carvey, William J. Daily and Larry R. Dennison for Apparatus and Methods for Connecting Modules Using Remote Switching.

Two important measures of the quality of a topology are average path length and maximum channel load. The average path length, D avg , of a network is the average number of channels that must be traversed to send a packet between two randomly selected nodes in a network. In the radix-8, two-dimensional torus network of FIG. 1 , the average packet traverses two channels in each dimension, giving P avg =4. In general, the average path length in a k-ary n-cube is kn/4 (for k even) and n(k/4-1/4k) (for k odd) while the average path length in a k-ary n-mesh is n(k/3-1/3k). Average path length is a good predictor of network latency. The lower the average path length of a topology, the lower is the latency of a network using that topology, assuming the latency of an individual channel remains constant.

The maximum channel load of a network, C max , is the largest number of packets that traverse a single channel when all nodes send packets to all other nodes divided by the number of nodes squared. In the 8-ary 2-cube network of FIG. 1 , for example, the channel load is uniform with C max =C i =1 for every channel i. In general, the maximum channel load in a k-ary n-cube is k/8 while the maximum channel load in a k-ary n-mesh is k/4. Maximum channel load is a good predictor of the throughput of a network. The lower the maximum channel load of a topology, the higher is the throughput of a network using that topology, assuming the bandwidth of an individual channel remains constant.

SUMMARY OF THE INVENTION

While the folded torus network offers a method of packaging a regular k-ary n-cube network so that all of the channels remain short, it is inherently inefficient because nodes that are packaged very close to one another can communicate only via circuitous routes. For example, in FIG. 3 , nodes ( 2 , 0 ) and ( 7 , 0 ) are on the same backplane. A message from ( 2 , 0 ) to ( 7 , 0 ), however, must leave this backplane and traverse five backplane-backplane connections before arriving back at the backplane. The problem becomes even more acute as the x-dimension grows in radix. A more efficient topology would minimize this off-backplane communication by allowing direct communication between nodes on the same backplane. A more efficient topology would also provide good performance without needing an end-around connection that must be removed for expansion.

One mechanism for increasing system performance in terms of average path length and maximum channel load and for making a network less susceptible to loss of end around connections is to add one or more dimensions to the network. However, adding a dimension increases the node degree by two, thus increasing the complexity of each node.

In accordance with one aspect of the invention, benefits of an added dimension without the increased node degree can be obtained by connecting one subset of channels within the network in one dimension and connecting another subset of the network channels in a different dimension. By distributing the nodes connected by the two channel subsets throughout the network, messages transferred through the network always have ready access to each dimension, but since each node need only be connected to one dimension or the other, there is no increase in degree. The concept of interleaving differing connections along one or more dimensions can be extended to other configurations for improved performance. Further, connections can be modified to assure that all nodes on a backplane, for example, are connected in a single cycle. The increased path lengths between some nodes resulting from the addition of such local connections can generally be masked by alternating the configurations from backplane to backplane.

In accordance with the invention, a plurality of channels connect a plurality of nodes in each connection network. A first subset of the channels connects the nodes is in a first dimension. A second subset of the channels connects the nodes in at least a second dimension, and connection of the second subset of channels differs at different coordinates along the first dimension.

In a preferred embodiment, the second subset of the channels connects the nodes in cycles or linear arrays in a second dimension and the ordering of the nodes in cycles or linear arrays is different at different coordinates along the first dimension. A connection of the second subset of channels at a first set of coordinates in the first dimension connects nodes whose coordinates in the second dimension differ by a first number. The connection of the second subset of channels in a second set of coordinates in the first dimension connects nodes whose coordinates in the second dimension differ by a second number. Thus, for example, in even numbered positions along the x-dimension, adjacent nodes in the backplane may be connected. However, in odd numbered positions, every other node may be connected.

Preferably, cycles or linear arrays of the second subset of the channels connect multiple nodes of each of plural cycles or linear arrays in the first dimension. For example, where the channels of the first dimension are folded, the second subset of the channels may connect nodes on opposite sides of the folds, thus reducing loading in the first dimension.

A first set of coordinates in the first dimension of the second subset of channels may connect the nodes into a cycle or linear array along a second dimension while, at a second set of coordinates in the first dimension, the second subset of channels connect the nodes into a cycle or linear array along a third dimension.

A preferred embodiment of the invention is a three-dimensional array. The second subset of the channels includes channel connecting nodes in a second dimension and channels connecting nodes in a third dimension. The connections in the third dimension differ at different coordinates of the first and second dimensions.

Nodes connected in at least the second dimension at a coordinate along the first dimension may be packaged in a single packaging module, such as on a backplane or other circuit board.

BRIEF DESCRIPTION OF THE DRAWINGS

The foregoing and other objects, features and advantages of the invention will be apparent from the following more particular description of preferred embodiments of the invention, as illustrated in the accompanying drawings in which like reference characters refer to the same parts throughout the different views. The drawings are not necessarily to scale, emphasis instead being placed upon illustrating the principles of the invention.

FIG. 1 is an illustration of a prior art 8-ary 2 cube.

FIG. 2 is an illustration of a 8-ary 2 mesh.

FIG. 3 is an illustration of a prior art folded torus packaged onto five backplanes.

FIG. 4 is an alternating 3 cube embodying the present invention packaged onto five backplanes.

FIG. 5 is the alternating 3 cube of FIG. 4 with alternate node labeling showing the 3 cube nature of the network.

FIG. 6 is an illustration of an interleaved cube topology embodying the present invention.

FIGS. 7A and 7B illustrate the y-dimension connections for the interleaved cube of FIG. 6 in the even and odd backplanes, respectively.

FIG. 8 illustrates a prior art 20-node backplane for a doubly folded 3 cube.

FIG. 9 illustrates one z plane of a prior art doubly folded torus showing connection between four backplanes.

FIG. 10 illustrates an even numbered backplane for a three-dimensional interleaved torus embodying the invention.

FIG. 11 shows the z connections for the even numbered backplane in the three-dimensional interleaved torus.

FIG. 12 illustrates the z connections for an odd numbered backplane in the three-dimensional interleaved torus.

FIG. 13 illustrates the z connections for an even numbered backplane in an alternative three-dimensional interleaved torus.

FIG. 14 illustrates the z connections for an even numbered backplane in a second alternative to the three-dimensional interleaved torus.

DETAILED DESCRIPTION OF THE INVENTION

A network that allows more local communication between nodes that are physically close to one another is shown in FIG. 4 . In this network even-numbered backplanes E are wired identically to the backplanes of FIG. 3 while odd-numbered backplanes O connect the nodes on opposite sides of each x-loop that shares the backplane. This arrangement allows a node to access any other node on its backplane in at most six hops and crossing at most two backplane connectors.

The topology of FIG. 4 is best thought of as a three-dimensional network. The x-dimension selects the backplane, the y-dimension selects a pair of nodes within the backplane (as before), and the z-dimension selects the node within the pair. FIG. 5 shows the same network with the nodes renumbered to correspond to this three-dimensional interpretation of the network. Every node in this network has connections to its neighbors in the x-direction while only ‘even’ nodes have connections in the y-direction and only ‘odd’ nodes have connections in the z-direction. Because the connections alternate between y- and z- we refer to this network as an alternating cube. This particular alternating cube has a radix-5 mesh in the x-dimension, a radix-4 torus in the y-dimension, and a radix-2 torus in the z-dimension. While the end-around connections form a 10-cycle in the x-dimension, we ignore those connections for purposes of analyzing this topology and consider the 10-cycle as two linear arrays of 5-nodes at different coordinates in the z-dimension. One skilled in the art will understand that this topology can be generalized to any number of dimensions with any radix in each dimension and either mesh or torus end connections in each dimension.

The alternating 3-cube of FIGS. 4 and 5 has two advantages compared to the folded 2-cube of FIG. 3 and one significant disadvantage. First, it has a slightly shorter average path length (3.26 vs. 3.5). This reflects the utility of the connections on the odd backplanes. More importantly, this network works nearly as well with one or both of the end-around connections in the x-dimension omitted. When just one of the connections is omitted, for example, the average path length increases slightly from 3.26 to 3.30 and the maximum channel load remains constant. In contrast, the average path length of the torus network increases from 3.5 to 4.3 and the maximum channel load is doubled (from 1.25 to 2.5) when this connection is opened. Thus, a network with an alternating cube topology can omit the costly hardware required to reconfigure the right end-around channel to allow expansion. With a torus network, on the other hand, this hardware is strictly required.

The disadvantage of the alternating cube is its throughput. The maximum channel loading of the network of FIG. 4 is 1.33 vs. 1.25 for the torus of FIG. 3 . This 6% increase in maximum channel load corresponds to a 6% decrease in peak network throughput.

The increased C max of the alternating cube network is caused by a load imbalance on channels in the x-dimension. This imbalance can be corrected by connecting the even- and odd-numbered backplanes in an interleaved cube topology as shown in FIG. 6 . Because each node on a backplane can be reached by another node on the backplane without crossing to another backplane, loading of x-channels is reduced. This configuration has both a lower average path length (2.87 vs. 3.5) and a lower maximum channel load (1.20 vs. 1.25) than the torus network. Like the alternating cube, this network has nearly the same performance (D avg =2.93, C max =1.20) when one of the end-around connections is omitted to save the cost of reconfiguration.

In the topology of FIG. 6 , the nodes in each even-numbered backplane are wired in a single cycle that we use to define the y-dimension of the network, rather than the two cycles of FIG. 5 . All nodes in FIG. 6 are labeled with a two-digit address where the first digit gives the backplane number (x-coordinate) and the second digit gives the y-coordinate. Moving in the x-dimension changes only the x-coordinate while moving in the y-dimension changes only the y-coordinate. The nodes in each odd-numbered backplane of FIG. 6 are also connected in a single cycle, but not the same cycle as the even-numbered backplanes.

The relation between the even-numbered and odd-numbered backplanes in the interleaved cube is illustrated in FIGS. 7A and B . In the even-numbered backplanes of FIG. 7A , each node is connected to the next node ahead and the next node behind along the cycle. Nodes in the odd-numbered backplanes of FIG. 7B , on the other hand, are connected to the nodes three hops ahead and behind along the cycle.

It is significant that the even and odd backplanes support different configurations. The single cycle of each backplane enables communications between any two nodes on a backplane without leaving the backplane. However, the single cycle does result in increased distances between some nodes. For example, when node 001 is connected directly to node 031 in FIG. 5 , the corresponding node 04 in FIG. 6 is now three hops away from node 07 . However, if one considers a message traveling across many backplanes in a large system, a message may be routed from the lower torus to the upper torus in a single hop from the node 14 to the node 17 . Whether a message is routed to a different x-dimension torus in either an even or an odd backplane is determined by which type of backplane provides the shortest path.

Up to this point we have described the alternating cube and interleaved cube topologies in two dimensions for simplicity. One skilled in the art will understand that these topologies can be realized with any number of dimensions.

For example, in a preferred embodiment of the invention, the interleaved cube topology is applied to a 3-dimensional doubly-folded torus network (a doubly-folded 3-cube). A single backplane of the unmodified torus network, shown in FIG. 8 , contains 20 nodes in four cycles of five. Each node is labeled with a three-digit number giving its relative x, y, and z-coordinates within the backplane. There are six ports on each node. The positive and negative ports in the x-dimension connect to the right and left edges of the backplane respectively. Similarly the positive and negative ports in the y-dimension connect to the top and bottom edges of the backplane. In contrast, the positive and negative ports in the z-dimension connect each node to the other nodes in its cycle. The z-dimension is wired entirely within the backplane while the x- and y-dimensions are wired entirely with backplane to backplane connections.

The x- and y-connections between the backplanes of the doubly-folded 3-cube network are illustrated in FIG. 9 . The figure shows one of the five z-planes for a 4×4×5 3-cube. The nodes in the other four z-planes are wired identically. The figure illustrates how the network is folded in both the x- and y-dimensions to realize a torus connection in both dimensions using only connections to nodes on adjacent backplanes. Note that since FIG. 9 has been extended to four backplanes, the numbering of the nodes does not directly correspond to that of FIG. 8 . However, if one considers the backplane of FIG. 8 to be the upper left backplane of FIG. 9 , the nodes 00 z, 03 z, 30 z and 33 z of FIG. 9 correspond, for example, to nodes 000, 010, 100 and 110 in one z plane or node 001, 011, 101 and 111 in another z plane.

FIG. 9 can be compared to two backplanes of FIG. 3 . The difference between the circuits is that, in the y-dimension, the nodes of FIG. 9 are folded in the same manner as in the x-dimension. This allows for shorter end around channels and expansion in the y-dimension through edge connection of backplanes.

While FIG. 9 shows a 4×4×5 torus, one skilled in the art will understand how the network can be expanded at the right edge and the top edge to produce a network with any even number of nodes in the x- and y-dimensions and 5-nodes in the z-dimension. In the preferred embodiment the network may be expanded to 560-nodes with a 14×8×5 torus topology.

The doubly-folded 3-cube network is converted to a 3-dimensional interleaved cube by connecting the even-numbered backplanes as illustrated in FIGS. 10 and 11 and the odd-numbered backplanes as illustrated in FIG. 12 . In the doubly-folded torus network, an even-numbered backplane is one in which the sum of the x- and y-coordinates of the backplane is even. The even- and odd-numbered backplanes alternate in a checkerboard manner as the network is expanded.

As shown in FIG. 10 , the nodes of each even-numbered backplane are connected in a single 20-cycle. FIG. 11 shows this same 20-cycle but with the x- and y-dimension channels omitted for clarity. Advancing along the z-channels of the network moves a packet through the 5-nodes in the z-dimension, but then, rather than looping back as in FIG. 8 , the next hop in the z-dimension jumps to another side of an x- or y-torus. Referring to FIG. 9 , the single cycle of FIG. 10 would connect node 00 z not only to the other four nodes in its z cycle of FIG. 8 but also to the other z cycles of FIG. 8 . For example, in FIG. 9 , node 00 z also connects to node 03 z on the opposite side of the y-torus fold, node 30 z on the opposite side of the x-torus fold and to node 33 z on completely different x- and y-tori. As described above in the case of the 2-D interleaved cube network, this connection facilitates communication between nodes that are physically close to one another even though they may be at opposite positions along the x- or y-torus network.

The connections between the nodes on the odd-numbered backplanes for the 3-dimensional interleaved torus are shown in FIG. 12 . As with the even-numbered backplanes, the nodes are connected in a 20-cycle. However, it is a different 20-cycle than in the even-numbered backplanes. As seen by a comparison of FIGS. 11 and 12 , while advancing along a z-channel in an even-numbered backplane ( FIG. 11 ) moves one step forward or backward along the z-cycle, advancing over a z-channel in an odd-numbered backplane ( FIG. 12 ) moves three steps forward or backward along the cycle.

Somewhat better performance in channel loading can be achieved by connecting the even backplanes of the interleaved 3-D torus network in two 10-cycles as shown in FIG. 13 . With this connection, advancing along a z-channel moves two steps forward or backward along the z-cycle. However, because the step size, two, is not relatively prime to the length of the cycle, 20, the connection does not form a single 20-cycle, but rather forms two 10-cycles. For applications where it is important that all of the nodes be connected in a single 20-cycle in the z-dimension, this topology can be modified as illustrated in FIG. 14 . In this figure the two 10-cycles of FIG. 13 have been fused by connecting nodes 004 and 010 and nodes 003 and 011 . The irregular topology of FIG. 14 has slightly lower performance than the topology of FIG. 13 but has the advantage that a single backplane is completely connected without any external end-around connections. However, the irregularity complicates the computation of routes.

One skilled in the art will understand that there are many variations on the interleaved topology described here. In particular there may be any number of dimensions (not just 2 or 3) and each dimension may be organized as a torus or a mesh or even a more complex topology such as a butterfly or tree. The network need not be packaged on backplanes but may use some other packaging technology and hierarchy. For example, the nodes may be directly cabled together. The last dimension of the network need not be strictly local to a single backplane. The last dimension of the network need not be organized into a single cycle, but rather could be arranged as a group of cycles or linear arrays, or some combination of the two. Also, the cycles in the odd-numbered backplanes need not always advance by three relative to the even numbered backplanes. They can advance by any number that is relatively prime to the radix of the cycle and can even advance by irregular amounts.

While this invention has been particularly shown and described with references to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims

1. An interconnection network comprising:a plurality of nodes; and a plurality of channels connecting the nodes: a first subset of the channels connecting the nodes in a first dimension; and a second subset of the channels connecting the nodes in cycles or linear arrays in a second dimension, the ordering of the nodes in the cycles or linear arrays being different at different coordinates along the first dimension; the connection of the second subset of channels at a first set of even-numbered coordinates in the first dimension connecting nodes whose coordinates in the second dimension differ by a first number; and the connection of the second subset of channels at a second set of odd-numbered coordinates in the first dimension connecting nodes whose coordinates in the second dimension differ by a second number.

2. An interconnection network as claimed in claim 1 wherein;at a first set of coordinates in the first dimension the second subset of channels connect the nodes into a cycle or linear array along a second dimension; and at a second set of coordinates in the first dimension the second subset of channels connect the nodes into a cycle or linear array along a third dimension.

3. An interconnection network as claimed in claim 1 wherein the network connects the nodes of a parallel computer.

4. An interconnection network as claimed in claim 1 wherein the network connects line cards in a network router.

5. An interconnection network as claimed in claim 1 wherein the nodes connected in at least the second dimension at a coordinate along the first dimension are packaged in a single packaging module.

6. An interconnection network as claimed in claim 5 wherein all nodes in the packaging module are connected in a single cycle within the module.

7. An interconnection network as claimed in claim 5 wherein the single packaging module is defined by a circuit board.

8. An interconnection network as claimed in 5 wherein the single packaging module is defined by a backplane.

9. An interconnection network as claimed in claim 1 wherein the first subset of channels forms cycles or linear arrays that are folded.

10. An interconnection network as claimed in claim 1 wherein all nodes in a packaging module are connected in a single cycle within the module.

11. An interconnection network as claimed in claim 1 wherein the second subset of the channels includes channels connecting nodes in a second dimension and channels connecting nodes in a third dimension, the connections in the third dimensions differing at different coordinates of the first and second dimensions.

12. An interconnection network as claimed in claim 1 wherein channels of the second subset form cycles, each of which intersects each array of the first subset of channels at multiple nodes.

13. An interconnection network as claimed in claim 1 wherein each of the plurality of channels directly connects a distinct pair of the nodes.

14. An interconnection network comprising:a plurality of nodes; and a plurality of channels connecting the nodes: a first subset of the channels connecting the nodes in a first dimension; and a second subset of the channels connecting the nodes in cycles or linear arrays in a second dimension, the ordering of the nodes in the cycles or linear arrays being different at different coordinates along the first dimension, the cycles or linear arrays of the second subset of the channels connecting multiple nodes of each of plural cycles or linear arrays in the first dimension, the first subset of the channels being cycles or linear arrays that are folded and the second subset of the channels connecting nodes on opposite sides of the folds.

15. An interconnection network as claimed in claimed 14 wherein each of the plurality of channels directly connects a distinct pair of the nodes.

16. An interconnection network comprisinga plurality of nodes; and a plurality of channels connecting the nodes: a first subset of the channels connecting the nodes in a first dimension; a second subset of the channels connecting the nodes in a second dimension; a third subset of the channels connects the nodes in cycles or linear arrays in a third dimension, the ordering of the nodes in the cycles or linear arrays being different at different coordinates along the first and second dimensions; the connection of the third subset of channels at a first set of coordinates in the first and second dimensions connecting nodes whose coordinates in the third dimension differ by a first number; the connection of the third subset of channels at a second set of coordinates in the first and second dimensions connecting nodes whose coordinates in the third dimension differ by a second number; and the first set of coordinates corresponding to locations where the sum of the coordinates in the first and second dimensions is even.

17. An interconnection network as claimed in claim 16 wherein each of the plurality of channels directly connects a distinct pair of the nodes.

18. A method of connecting nodes in the network comprising:connecting the nodes in a first dimension with a first subset of channels; and connecting the nodes in at least a second dimension with a second subset of channels, the second subset of channels connecting the nodes in cycles or linear arrays in a second dimension, the ordering of the nodes in the cycles or linear arrays being different at different coordinates along the first dimension; the connection of the second subset of channels at a first set of coordinates in the first dimension corresponding to even-numbered positions connecting nodes whose coordinates in the second dimension differ by a first number; and the connection of the second subset of channels at a second set of coordinates in the first dimension corresponding to odd-numbered positions, connecting nodes whose coordinates in the second dimension differ by a second number.

19. A method as claimed in claim 18 wherein the first subset of channels forms cycles or linear arrays that are folded.

20. A method as claimed in claim 18 wherein each channel of the first and second subsets of channels directly connects a distinct pair of the nodes.

21. A method of connecting nodes in the network comprising:connecting the nodes in a first dimension with a first subset of channels; and connecting the nodes in at least a second dimension with a second subset of channels, the second subset of channels connecting the nodes in cycles or linear arrays in a second dimension, the ordering of the nodes in the cycles or linear arrays being different at different coordinates along the first dimension; the cycles or linear arrays of the second subset of the channels connecting multiple nodes of each of plural cycles or linear arrays in the first dimension; the first subset of the channels being cycles or linear arrays that are folded and the second subset of the channels connecting nodes on opposite sides of the folds.

22. A method as claimed in claim 21 wherein:at a first set of coordinates in the first dimension the second subset of channels connect the nodes into a cycle or linear array along a second dimension; and at a second set of coordinates in the first dimension the second subset of channels connect the nodes into a cycle or linear array along a third dimension.

23. A method as claimed in claim 21 wherein the network connects the nodes of a parallel computer.

24. A method as claimed in claim 21 wherein the network connects line cards in a network router.

25. A method as claimed in claim 21 wherein the nodes connected in at least the second dimension at a coordinate along the first dimension are packaged in a single packaging module.

26. A method as claimed in claim 25 wherein all nodes in the packaging module are connected in a single cycle within the module.

27. A method as claim 25 wherein the single packaging module is defined by a circuit board.

28. A method as claimed in claim 25 wherein the single packaging module is defined by a backplane.

29. A method as claimed in claim 21 wherein all nodes in a packaging module are connected in a single cycle within the module.

30. A method as claimed in claimed on 21 wherein the second subset of the channels includes channels connecting nodes in a second dimension and channels connecting nodes in a third dimension, the connections in the third dimensions differing at different coordinates of the first and second dimensions.

31. An interconnection network as claimed in claim 21 wherein channels of the second subset form cycles, each of which intersects each array of the first subset of channels at multiple nodes.

32. A method as claimed in claim 21 wherein each channel of the first and second subsets of channels directly connects a distinct pair of the nodes.

33. A method of connecting nodes in the network comprising:connecting the nodes with a first subset of channels in a first dimension; connecting the nodes with a second subset of channels in a second dimension; and connecting the nodes with a third subset of channels in at least a third dimension, the third subset of the channels connecting the nodes in cycles or linear arrays being different at different coordinates along the first and second dimensions; the connection of the third subset of channels at a first set of coordinates in the first and second dimensions connecting nodes whose coordinates in the third dimension differ by a first number; the connection of the third subset of channels at a second set of coordinates in the first and second dimensions connecting nodes whose coordinates in the third dimension differ by a second number; and the first set of coordinates corresponding to locations where the sum of the coordinates in the first and second dimensions is even.

34. A method as claimed in claim 33 wherein each of the channels of the first, second and third subsets of channels directly connects a distinct pair of the nodes.