Interconnection network extendable bandwidth and method of transferring data therein

• US 6,304,568 B1
• Filed: 01/26/1998
• Issued: 10/16/2001
• Est. Priority Date: 01/27/1997
• Status: Expired due to Fees

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Kim, Kab-young
• Primary Examiner(s): Holloway, III, Edwin C.
• Application Number:
  US09/013,190
• Time in Patent Office: 1,359 Days
• Field of Search: 370/387, 370/388, 340/825.79, 340/825.8, 340/2.1, 340/2.6, 340/2.7, 709/238
• US Class Current: 370/387
• CPC Class Codes: H04L 45/06 Deflection routing, e.g. hot-potato routing

Abstract

An interconnection network in which a control plane and data planes are separate and the bandwidth is extendable, and a method for transferring data includes a control plane including control routers connected to the processors, for exchanging control information and generating information on setting a path for a message transmitted between the processors, and a control line for connecting the control routers together in a predetermined topology, and one or more data planes each including data routers each connected to the processors, for transmitting and receiving messages using the path information on messages generated by the control router, and a data transfer line for connecting the data routers together in the same topology as that of the control plane. It is possible to arbitrarily extend the bandwidth of only the data plane, since the control plane and the data plane are separate, and to construct a high speed interconnection network since the latency time consumed when the path is traversed is independent of the size and frequency of the data transferred across the interconnection network. Therefore, it is possible to construct a flexible multicomputer system, by taking advantage of the large bandwidth and short latency time.

Claims

1. An interconnection network for a multiprocessor communication system, said interconnection network comprising a plurality of nodes arranged in a predetermined topographical structure, each of said nodes comprising:

a single interface connected to a corresponding single processor for receiving source message data from said processor and transmitting destination message data to said processor;

a control router connected to receive control data from said interface;

a first data router connected to receive and transmit message data over a first data connection with said interface;

a second data router connected to receive and transmit message data over a second data connection with said interface;

a control line connected to each of said control routers of each of said nodes for transferring said control data from a control router of a source node to a control router of a destination node, said control line and said control routers forming a control plane arranged in said predetermined topographical structure;

a first data line connected to each of said first data routers of each of said nodes for transferring message data from a first data router of said source node to a first data router of said destination node, said first data line and said first data routers forming a first data plane arranged in said predetermined topographical structure; and

a second data line connected to each of said second data routers of each of said nodes for transferring message data from a second data router of said source node to a second data router of said destination node, said second data line and said second data routers forming a second data plane arranged in said predetermined topographical structure, wherein said first and second data lines carry only message data and said control line carries only control data.

2. The interconnection network for a multiprocessor communication system as set forth in claim 1, wherein each said interface of each of said nodes comprises further data connections for connecting to further data routers of said node to extend a bandwidth of said interconnection network.

3. The interconnection network for a multiprocessor communication system as set forth in claim 1, wherein predetermined topographical structure is ring, a Banyan, a tree, a hypercube, a mesh or a torus interconnection network.

4. The interconnection network for a multiprocessor communication system as set forth in claim 1, wherein said control data comprises header information including an address of said destination node, and said control router of said source node utilizes said address to set a message data communication path through intermediate nodes to said destination node over one of said first and second data planes and to set a control data communication path through said intermediate nodes to said destination node over said control plane.

5. The interconnection network for a multiprocessor communication system as set forth in claim 4, wherein each of said control routers of said intermediate nodes in said communication path determine whether a data router in said message data communication path is free to transmit message data, such that the control router, which determines that a corresponding data router is not free, blocks the transmission of said message data until it is determined that said data router is free, said message data communication path remaining set until said message data is transmitted through said data router.

6. A method of transferring data in an interconnection network comprising a plurality of nodes arranged in a predetermined topographical structure, each of said nodes comprising a single processor for transmitting and receiving control data and message data, a control router disposed in a control plane, one or more data routers, and one or more data planes in each of which one of the data routers is disposed, respectively, said method comprising the steps of:

detecting header information in control data transmitted from a source processor in a source node and determining destination processor a destination node in response to address data in said header information;

setting a control data communication path through intermediate nodes to said destination node over said control plane and setting a separate message data communication path through said same intermediate nodes to said destination node over one of said data planes;

transferring only said control data from the control router of said source node to each control router in said control data communication path;

determining, by each of said control routers in said control data communication path, whether a corresponding data router in said message data communication path is free to transmit message data through the correspond node;

wherein said message data communication path transmits only message data;

blocking the message data communication path to prevent transmission of said message data, when it is determined that one of said data routers is not free, while maintaining said control data communication path;

releasing the message data communication path, when it is determined that said one of said data routers is free, to transmit said message data to the processor of the destination node.

7. A method of transferring data in an interconnection network comprising a plurality of nodes arranged in a predetermined topographical structure, each of said nodes comprising a single processor for transmitting and receiving control data and message data, a control router disposed in a control plane, a first data router disposed in a first data plane and a second data router disposed in a second data plane, said method comprising the steps of:

detecting header information in control data transmitted from a source processor in a source node and determining destination processor a destination node in response to address data in said header information;

setting a control data communication path through intermediate nodes to said destination node over said control plane and setting a separate message data communication path through said same intermediate nodes to said destination node over one of said first and second data planes;

transferring only said control data from the control router of said source node to each control router in said control data communication path;

determining, by each of said control routers in said control data communication path, whether a corresponding data router in said message data communication path is free to transmit message data through the correspond node;

wherein said message data communication path transmits only message data;

blocking the message data communication path to prevent transmission of said message data, when it is determined that one of said data routers is not free, while maintaining said control data communication path;

releasing the message data communication path, when it is determined that said one of said data routers is free, to transmit said message data to the processor of the destination node.