Method and apparatus for routing packets in a multinode computer interconnect network

• US 4,742,511 A
• Filed: 06/13/1985
• Issued: 05/03/1988
• Est. Priority Date: 06/13/1985
• Status: Expired due to Term

• Current Assignee: Texas Instruments, Inc.
• Original Assignee: Texas Instruments, Inc.
• Inventors: Johnson, Douglas A.
• Primary Examiner(s): Olms, Douglas W.
• Application Number:
  US06/744,583
• Time in Patent Office: 1,055 Days
• Field of Search: 370/94, 370/86, 370/60, 370/89, 370/88
• US Class Current: 370/406
• CPC Class Codes: G06F 15/17337 Direct connection machines, e.g. completely connected computers, point to point communication networks coupling between buses G06F13/4004
• CPC Class Codes: H04L 12/4637 Interconnected ring systems
• CPC Class Codes: H04L 45/06 Deflection routing, e.g. hot-potato routing

Abstract

A method for routing user packets in a multinode interconnect network is disclosed. Each packet has a destination node (10). Each node (10) is connected by input and output lines (12) and has a plurality of packet storage links (34) and (56-60). Each link (34) and (56-60) is connected to an input line (18-24) and has a number of packet storage buffers (38-44) and is connectable to any of the node'"'"'s output lines (26-32). During operation node (10) uses a set of routing instructions and a look-up table (192) to determine which user packets are sent out on which output lines (26-32). The routing rules determine which output line is connected to the shortest path to the destination node, and ensure that the node (10) will always have an empty storage buffer (38-44) for each input line (18-24) for receiving further user packets.

Claims

1. A method for routing user packets in an interconnect network, the interconnect network having a plurality on nodes each having a plurality of output lines and packet storage, each user packet having a destination node, the method comprising the steps of:

receiving a number of packets into a first node;

storing the packets;

sensing the destination node of at least one stored packet;

determining a preferred output line for the stored packet, the preferred output line being associated with the shortest nodal travel distance to the destination node, wherein each output line is associated with an input line, including the steps of;

originating a plurality of node distance packets at each node in the network, each node distance packet including a node identity and a distance number originally set to a base value,

transmitting the node distance packets from each node to connected nodes,

receiving the node distance packets by each node from connected nodes via input lines,

at each node, incrementing the distance number of each received node distance packet by a constant,

setting a value of a corresponding nodal distance array member to the distance number from each incremented distance packet, the array member being referred by the node identity and the output line associated with the input line on which the distance packet was received,

retransmitting from each node to connected nodes the incremented distance packets,

receiving the incremented distance packets by each node,

at each node, further incrementing the distance number of each received incremented distance packet,

at each node, comparing the further incremented distance number of each distance packet to the value of a corresponding array member,

discarding any further incremented distance packet whose distance number equals or exceeds a nonzero value of the corresponding array member to the distance number,

at each node, retransmitting the nondiscarded distance packets to all connected nodes,

repeating the incrementing, comparing, discarding, setting and retransmitting steps until no nondiscarded node distance packets remain;

choosing the array member having the shortest nodal travel distance from all array members referenced by the destination node;

setting the preferred output line of the packet to the output line referencing the chosen array member; and

transmitting the packet on its determined preferred output line.

2. The method of claim 1, further including the steps of:

during operation, sensing by a node of a failure by a connected node to receive a user packet;

erasing all present values in the nodal distance array of the sensing node in response to the failure;

sending a fault packet by the sensing node to all connected nodes in response to the failure;

erasing the nodal distance array of each node receiving a fault packet;

retransmitting the fault packet by each receiving node to further connected nodes and erasing the nodal distance arrays of the further connected nodes in response to fault packet receipt, the process continuing until all nodal distance arrays in the network have been erased; and

regenerating the nodal distance array of each node by generating, receiving, incrementing, comparing, discarding, setting and retransmitting node distance packets in a manner similar to originating the nodal distance arrays.

3. A method for routing user packets in an interconnect network, the interconnect network having a plurality of nodes each having a plurality of output lines and packet storage, each user packet having a destination node, wherein each node has a plurality of packet storage links, each having a plurality of packet storage buffers, an input line being connected to each link, the method comprising the steps of;

receiving a number of packets into a first node;

storing each received packet in a buffer;

sensing the destination node of at least some stored packets;

determining a preferred output line and an associated nodal travel distance to a packet,s destination node for each stored packet;

for each link having at least one nonempty buffer, selecting a packet having the shortest nodal travel distance of all the packets stored in the link;

assigning the selected packets, in increasing order of nodal travel distance, to their preferred output lines, if the preferred output line for a first packet has not already been assigned to another first packet;

after assigning the selected packets, identifying which links would still be full after transmission of the assigned selected packets if any output lines remain unassigned;

for each prospective full link, if any output lines remain unassigned, assigning a packet stored in the link to an unassigned line until the unassigned output lines are exausted or until no prospective full link remains; and

transmitting ones of the portion of stored packets on their respective preferred output lines.

4. The method of claim 3 further including the step of:

if any output lines remain unassigned after the step of assigning packets from prospective full links, assigning unassigned packets to their respective preferred output lines if available, until no unassigned preferred output lines remain or until each link has an assigned packet.

5. The method of claim 4, further including the steps of:

after identifying the prospective full links, assigning unassigned output lines to second packets stored in the prospective full links, each assigned second packet having a preferred output line equal to an unassigned line; and

identifying which links would remain full after transmission of the selected and second packets.

6. An interconnect network for transmission of user packets including a plurality of nodes, each node being connected by input and output lines to a plurality of other nodes, each node comprising:

means for receiving a plurality of user packets each having a destination node;

means for storing received packets;

means for sensing the destination node of each stored packet;

means for determining a preferred output line for each stored packet, said preferred output line being connected to a path having the shortest nodal travel distance to said storage packet'"'"'s destination node;

means for assigning stored packets to their respective preferrred output lines, said means selecting packets for assignment according to the increasing order of the packets, respective nodal travel distances; and

means for transmitting said packet on its preferred output line;

wherein said sensing means, determining means and assigning means comprise processor means, said processor means including memory means for storing a set of packet routing rules, and wherein said node includes a plurality of links each comprising a plurality of said packet storage means, said routing rules, after selecting user packets for transmission according to their nodal travel distances and availability of preferred output lines, selecting packets for transmission on and remaining available output lines according to whether each said link prospectively has at least one empty packet storage means, said routing rules preferentially selecting one or more stored packets within prospectively nonempty links for transmission on the remaining available output lines.

7. The network of claim 6, wherein:

said memory means further includes a look-up table giving the nodal travel distance to any node as measured from any output line;

said links each communicating to an input line, each packet storage means comprising a buffer capable of storing a user packet, each buffer being selectively connectable to an output line;

said routing rules assigning user packets stored in said buffers in a plurality of steps including a first step and a second step;

said routing rules in said first step selecting, for each link storing one or more user packets, one of the user packets having the shortest nodal travel distance to comprise a first set of packets;

said routing rules further in said first step assigning, for each packet in said first link, in increasing order of said packets'"'"' nodal travel distances, a preferred output line, unless said output line has already been assigned to a packet in said first set; and

said processor means in said second step inspecting each link to determine whether all buffers in any link are full, said routing rules assigning one packet from a full link to any available output line until there are no more full links or until all output lines have been assigned to packets.

8. The network of claim 7, wherein:

said look-up tables of said nodes are initialized at the beginning of operation, each said node generating its look-up table based on received distance packets originating from each other node in the network, each distance packet comprising a node identification identifying the originating node and a nodal distance number that the distance packet has travelled between its originating node and the receiving node.

9. The network of claim 8, wherein

each node, during initialization of said look-up tables, originates a plurality of said distance packets, each of said distance packets originally having a nodal distance number of zero, each node transmitting one distance packet on each output line to connected nodes;

each node receiving distance packets on its input lines, each input line being paired with an output line, the pair connecting the same nodes;

the distance number of each said received distance packet being incremented by one by said receiving node;

said receiving node next comparing said incremented distance number to a numerical entry in said look-up table, said entry being referenced by the output line paired with the input line on which said distance packet appeared and said packet'"'"'s node identification;

said receiving node discarding said distance packet if said packet'"'"'s incremented distance number is not less than said entry where said entry is nonzero; and

said receiving node, if said incremented distance number is less than said entry or said entry is zero, setting said entry to said distance number, said receiving node then replicating said distance packet and retransmitting said distance packet on all output lines to neighboring nodes.

10. The network of claim 9, wherein:

each node is capable of sensing a failure of connected node to receive a packet transmitted to said connected node by said sensing node;

said sensing node erasing its look-up table in response to sensing said failure;

said sensing node transmitting a fault packet on each of its output lines to connected nodes in response to said failure;

said connected nodes erasing their look-up tables and likewise retransmitting said fault packets, until all nodes in said network have erased their look-up tables; and

each node reconstructing its erased look-up table in a manner similar to the original generation of said look-up table.

11. The network of claim 7, wherein:

each said link is connected to an interior output line, said interior output lines each connecting to a crossbar, said crossbar terminating a plurality of exterior output lines; and

said processor means selecting up to one packet stored in a link to communicate to said crossbar on said link'"'"'s interior output line, said processor means directing said crossbar means to communicate said selected packet to one exterior output line.

12. The multinode interconnect network of claim 11, wherein:

said exterior output lines include a plurality of output lines connected to other nodes and a local output line to a local device interface;

said processor means directing said crossbar means to communicate up to one user packet stored in a link with said local output line, said packet having a destination node equal to said packet'"'"'s present location; and

said links being selectively connected to also receiving a local packet input path from said device interface, said processor means selecting up to one buffer for receiving and storing a local packet.

13. The network of claim 11, wherein:

each said link includes a demultiplexer selecting one of said buffers for receipt of a packet in response to a command from said processor means, said demultiplexer having one of said input lines connected to its input;

each said link further including a multiplexer selecting one of said buffers for transmission of a packet stored in said buffer to said crossbar in response to a command from said processor means;

said multiplexer having an output terminal connected to an interior output line; and

said processor means being connected to each said buffer so as to be capable of reading data from and writing data into said buffer.