Synchronization methods for distributed processing systems having replicated data

• US 6,021,118 A
• Filed: 02/10/1997
• Issued: 02/01/2000
• Est. Priority Date: 02/10/1997
• Status: Expired due to Term

• Current Assignee: Alcatel-Lucent USA, Inc. (Nokia Corporation)
• Original Assignee: Lucent Technologies, Inc. (Nokia Corporation)
• Inventors: Leung, Kin K., Winkler, Peter M., Houck, David J.
• Primary Examiner(s): Pham, Chi H.
• Assistant Examiner(s): QURESHI, AFSAR M
• Application Number:
  US08/795,263
• Time in Patent Office: 1,086 Days
• Field of Search: 370/229, 370/231, 370/254, 370/255, 370/400, 370/503, 370/411, 395/200.78, 395/200.79, 395/200.62, 395/200.63, 395/200.64, 395/200.65, 395/200.68, 395/181.02, 395/182.02
• US Class Current: 370/254
• CPC Class Codes: G06F 15/17381 Two dimensional, e.g. mesh, torus

Abstract

A data synchronization system, which in one embodiment, uses a ShuffleNet topology requiring an even number, N=2m, of nodes in the system. These nodes are organized into two sets, X=x₀, . . . , x_m-1 and Y=y₀, . . . , y_m-1, wherein the subscripts are always to be taken modulo m. Each "round" of communication entails simultaneously synchronizing the nodes in X with nodes in Y according to a matching between the two sets. The rounds are grouped into two "batches," batch B_j which consists of rounds R_2j-1 and R_2j-2 for j≧1. During each odd batch B_2j-1, each x_i synchronizes with y_2i+2j-2 and with y_2i+2j-1. In another embodiment, the data synchronization is based on a hypercube scheme, wherein each node is labeled by a binary string and any two nodes with their labels differing by one bit are connected by an edge and only adjacent nodes, i.e. those nodes connected by an edge, can communicate and exchange data directly according to an update schedule. In a third embodiment, a hypercube scheme is used, but the number of nodes is assumed to be a power of 2 or N=2^m. This embodiment, like the second embodiment, uses the labeling of nodes by their binary representation, but the matchings of nodes used to determine the update schedule is not confined to the hypercube edges. Instead, a general cyclic matching scheme is used.

Claims

1. A synchronization method for distributed processing systems having replicated data comprising the steps of:

establishing a network of computing nodes, each node having at least one data file to be shared with at least one other node;

implementing a ShuffleNet topology to control the flow of new data among the computing nodes, wherein there are an even number, N=2m, of nodes in the network;

organizing the computing nodes into two sets, X=x₀, . . . , x_m-1 and Y=y₀, . . . , y_m-1, wherein the subscripts are always to be taken modulo m and x and y are used as set designations;

synchronizing simultaneously the nodes in X with nodes in Y according to a matching between the two sets wherein rounds of communication are grouped into two batches.

2. The method of claim 1 wherein each batch consists of rounds, R_2j-1 R_2j for j≧

1, and in each odd batch, B_2j-1, each x_i synchronizes with y_2i+2j-2 and with y_2i+2j-1 and in even batches, B_2j, each y_i synchronizes with x_2i+2j-2 and x_2i+2j-1, i and j designate different integers corresponding to different nodes designations.

3. The method of claim 1 wherein a timer based schedule is utilized to synchronize the nodes in X with the nodes in Y.

4. The method of claim 1 wherein a precedence relationship based schedule is utilized to synchronize the nodes in X with the nodes in Y.

5. The method of claim 1 wherein a continuously repeating schedule is utilized to synchronize the nodes in X with the nodes in Y.

6. The method of claim 1 wherein the data files are global data files.

7. The method of claim 1 wherein the data files are a mixture of global and non-global data files.

8. A synchronization method for distributed processing systems having replicated data comprising the steps of:

establishing a network of computing nodes, each node having at least one data file to be shared with at least one other node;

implementing a hypercube topology to control the flow of new data among the computing nodes, wherein each node is labeled by a binary string and any two nodes which have their label differing by one bit are connected by an edge and only adjacent nodes can communicate and exchange data directly, wherein a timer based schedule is utilized to synchronize the computing nodes.

9. The method of claim 8 wherein edges in round, R_n, connect nodes with labels differing in their n^th bit numbered in predetermined fashion and each round of connectivity corresponds to one round of synchronization sessions.

10. The method of claim 8 further comprising the steps of:

labeling each of the N nodes by a binary string where the bits starting from the right are indexed by 0, 1, 2, . . . N;

establishing rounds of the connection topology by connecting nodes with labels differing only in the i-bit position at each round R_i where i=0, .left brkt-top.log₂ N.right brkt-top.-1;

establishing one round of updates where each pair of adjacent nodes in the corresponding round of connectivity synchronize data.

11. A synchronization method for distributed processing systems having replicated data comprising the steps of:

establishing a network of computing nodes, each node having at least one data file to be shared with at least one other node;

implementing a hypercube topology to control the flow of new data among the computing nodes, wherein each node is labeled by a binary string and any two nodes which have their label differing by one bit are connected by an edge and only adjacent nodes can communicate and exchange data directly, wherein a precedence relationship based schedule is utilized to synchronize the computing nodes.

12. A synchronization method for distributed processing systems having replicated data comprising the steps of:

establishing a network of computing nodes, each node having at least one data file to be shared with at least one other node;

implementing a hypercube topology to control the flow of new data among the computing nodes, wherein each node is labeled by a binary string and any two nodes which have their label differing by one bit are connected by an edge and only adjacent nodes can communicate and exchange data directly, wherein a continuously repeating schedule is utilized to synchronize the computing nodes.

13. A synchronization method for distributed processing systems having replicated data comprising the steps of:

establishing a network of computing nodes, each node having at least one data file to be shared with at least one other node;

implementing a generalized hypercube like topology to control the flow of data files among the computing nodes, wherein each node is labeled by a binary string and any two nodes labels x=(x₁, . . . , x_m) which have a non-zero binary vector and pairs are matched such that a matching of pairs, M_x, pairs a node labeled u with a node labeled x+u wherein the addition is coordinatewise and modulo 2, so that x+u+u=x and pair matching is symmetric.

14. The method of claim 13 wherein each pair of adjacent nodes in the topology can exchange data directly during a synchronized session at some point in time.

15. The method of claim 13 further comprising the step of cycling through matchings M_x(1) . . . , M_x(N-1), where the x(i)'"'"'s run through all the non-zero binary vectors of length m, thereby achieving a maximum update delay m if and only if for every i, the vectors x(i),x(i+1), . . . , x(i+m-1) with indices taken modulo n-1 are linearly independent over a predetermined function.

16. The method of claim 15 wherein the step of cycling utilizes a linear shift register.

17. The method of claim 13 wherein a non-global data files are updated without any intermediate nodes.

18. The method of claim 13 wherein a timer based schedule is utilized to synchronize the computing nodes.

19. The method of claim 13 wherein a precedence relationship based schedule is utilized to synchronize the computing nodes.

20. The method of claim 13 wherein a continuously repeating schedule is utilized to synchronize the computing nodes.