CN109842553B - Self-adaptive interconnection and routing control method and system for link resources - Google Patents
Self-adaptive interconnection and routing control method and system for link resources Download PDFInfo
- Publication number
- CN109842553B CN109842553B CN201711320765.3A CN201711320765A CN109842553B CN 109842553 B CN109842553 B CN 109842553B CN 201711320765 A CN201711320765 A CN 201711320765A CN 109842553 B CN109842553 B CN 109842553B
- Authority
- CN
- China
- Prior art keywords
- node
- nodes
- routing
- path
- communication
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention relates to a self-adaptive interconnection and routing control system facing direct network fast path link resources, which comprises a high-dimensional switch interconnection module, a routing information generation module, a routing table generation module, an interconnection method between high-dimensional switches and a corresponding deadlock-free routing method. The interconnection module of the high-dimensional switch can group the nodes according to the number of link resources of the fast path, set communication nodes, record grouping information and then interconnect the communication nodes of the high-dimensional switch by using the fast link. And the routing information generation module stores the quick path connection relation of the communication nodes of the high-dimensional switch through the mapping table. And the routing table generating module generates a routing forwarding table and a virtual channel switching table according to the grouping information and the mapping table, so that a deadlock-free routing function is realized.
Description
Technical Field
The invention belongs to the field of computers, and particularly relates to a parallel computer interconnection network, in particular to a routing algorithm of the parallel computer interconnection network.
Background
At present, in the field of network interconnection technology, how to utilize limited network switching resources to generate network performance with higher performance-to-cost ratio has become a bottleneck technology for network structure design and performance tuning. With the use of SDN configurable optical switching matrices in data centers and supercomputing clusters, flexible physical lightpath switching is increasingly being seen by network architects. It is worth noting that in the design of a network structure facing the next generation E-level supercomputing, a cross-dimension link of a direct network gradually becomes one of shortcut link policies for a network architect to accelerate network communication, since it can directly connect two switching nodes far away in the network through a physical link. Aiming at routing algorithms in networks, particularly Torus and Mesh networks, the related algorithms are focused on how to utilize the existing fixed network links, the network communication performance is improved by optimizing the routing algorithm, and meanwhile, the deadlock-free property of the routing is ensured.
The research of the ai-class (exaascale) high-performance computer is a current hotspot, and the system scale of the ai-class computer is expected to reach more than one hundred thousand node scale. The research of high-performance interconnection networks is a key ring for the implementation of the advanced computer. At present, a high-dimensional direct network is widely applied to a P-level (Peerscale) computer and has good performance.
With the expansion of network scale, the reduction of global communication performance of high-dimensional direct network becomes a difficulty in building ai-level system. In order to compensate for the short performance of direct network global communication, in the field of data centers, there have been studies to propose a scheme for increasing the fast path, and the use of this technology in high-performance computing networks is a future trend.
Disclosure of Invention
In order to solve the above problems, the present invention provides a method and a system for adaptive interconnection and routing control for direct network fast path link resources, wherein the method comprises:
a high-dimensional switch interconnection step, namely selecting a plurality of nodes of a high-dimensional switch as communication nodes, taking each communication node as a grouping identifier, and dividing all the nodes of the high-dimensional switch into a plurality of groups;
a routing information generation step, namely, when a quick path exists between any two communication nodes in all the communication nodes, generating a brother node information mapping table by using the quick path;
a step of generating a routing table, which is to establish a routing forwarding table and a virtual channel switching table of the high-dimensional switch by using the grouping information and the sibling node information mapping table;
and a route forwarding step, judging whether the route path forwarded by the information has the fast path or not according to the route forwarding table and the virtual channel switching table, so as to use the fast path or the traditional dimension order route path for routing.
The invention relates to a self-adaptive interconnection and routing control method, wherein the interconnection steps of a high-dimensional switch specifically comprise the following steps: setting a grouping parameter N of the high-dimensional switch, selecting N nodes with evenly distributed positions as N communication nodes, and evenly dividing all the nodes of the high-dimensional switch into N groups by taking each communication node as a center; wherein N is a positive integer.
The invention relates to a self-adaptive interconnection and routing control method, wherein the step of generating routing information specifically comprises the following steps: the node A and the node B in the communication nodes traverse all the communication nodes; if the node A and the node B are the same node at two ends of the fast path, the node A and the node B are brother nodes; and generating the sibling node information mapping table by all the quick paths.
The invention relates to a self-adaptive interconnection and routing control method, wherein the step of generating a routing table specifically comprises the following steps: and judging whether the fast path exists between the source node and the destination node by taking the communication node of the group where the routing path starting node is located as a source node and the communication node of the group where the routing path terminal node is located as a destination node, and routing through the traditional dimensional sequence routing path when only the traditional dimensional sequence routing path exists or the fast path exists and the step number of the fast path is greater than that of the traditional dimensional sequence routing path, otherwise, selecting the fast path for routing.
The invention relates to a self-adaptive interconnection and route control method, wherein the route forwarding step further comprises the following steps: a virtual channel switching step, dividing the fast path into three-stage routes, including a first stage from the initial node to the source node; a middle segment from the source node to the destination node; a terminal node from the destination node to the terminal node; and establishing a first virtual channel, a second virtual channel and a third virtual channel, wherein the first virtual channel and the second virtual channel are used in the initial segment, and the third virtual channel is used in the middle segment and the final segment.
The invention also relates to a direct network fast path link resource oriented self-adaptive interconnection and routing control system, which comprises:
the high-dimensional switch interconnection module is used for selecting a plurality of nodes of the high-dimensional switch as communication nodes, taking each communication node as a grouping identifier and dividing all the nodes of the high-dimensional switch into a plurality of groups;
a routing information generating module, configured to establish a sibling information mapping table of the high-dimensional switch, where when a fast path exists between any two nodes in all the communication nodes, the any two nodes are sibling nodes with each other, and the fast path is used to generate the sibling information mapping table;
a routing table generating module, configured to establish a routing forwarding table and a virtual channel switching table of the high-dimensional switch according to the packet information and the sibling node information mapping table;
and the route forwarding module is used for judging whether the route path forwarded by the information has the fast path or not according to the route forwarding table and the virtual channel switching table so as to use the fast path or the traditional dimension order route path for routing.
The invention relates to a self-adaptive interconnection and routing control system, wherein a high-dimensional switch interconnection module specifically comprises: setting a grouping parameter N of the high-dimensional switch, selecting N nodes with evenly distributed positions as N communication nodes, and evenly dividing all the nodes of the high-dimensional switch into N groups by taking each communication node as a center; wherein N is a positive integer.
The invention relates to a self-adaptive interconnection and routing control system, wherein a routing information generation module specifically comprises:
the traversal module is used for traversing all the communication nodes by the node A and the node B, and when the node A and the node B are nodes at two ends of the same fast path, the node A and the node B are brother nodes; wherein any one of the communication nodes is the node A, and any one of the communication nodes other than the node A is the node B;
and the mapping table module is used for generating the sibling node information mapping table from all the quick paths.
The invention relates to a self-adaptive interconnection and routing control system, wherein a routing table generating module specifically comprises: and judging whether the fast path exists between the source node and the destination node by taking the communication node of the group where the routing path starting node is located as a source node and the communication node of the group where the routing path terminal node is located as a destination node, and routing through the traditional dimensional sequence routing path when only the traditional dimensional sequence routing path exists or the fast path exists and the step number of the fast path is greater than that of the traditional dimensional sequence routing path, otherwise, selecting the fast path for routing.
The invention relates to a self-adaptive interconnection and route control method, wherein the route forwarding module further comprises: the virtual channel switching module is used for constructing a plurality of virtual channels to avoid routing deadlock; dividing the fast path into three-stage route including initial segment from the initial node to the source node; a middle segment from the source node to the destination node; a terminal node from the destination node to the terminal node; and establishing a first virtual channel, a second virtual channel and a third virtual channel, wherein the first virtual channel and the second virtual channel are used in the initial segment, and the third virtual channel is used in the middle segment and the final segment.
Drawings
FIG. 1 is a block diagram of an adaptive interconnection and routing control system according to the present invention.
Fig. 2 is a flow chart of a high-dimensional switch grouping method according to an embodiment of the invention.
Fig. 3 is a schematic diagram illustrating an example of implementation results of the high-dimensional switch grouping according to the present invention.
Fig. 4 is an exemplary diagram of an implementation process of the high-dimensional switch packet with N-4 according to the present invention.
Fig. 5 is a flowchart of an implementation of a method for generating routing information according to an embodiment of the present invention.
Fig. 6A is a schematic diagram illustrating an example of an implementation of the routing information generation method according to the present invention.
Fig. 6B is a sibling information mapping table diagram according to an embodiment of the routing information generation method of the present invention.
Fig. 7 is a flowchart of an implementation of a method for generating a routing forwarding table according to an embodiment of the present invention.
Fig. 8A is an exemplary diagram illustrating fast path new loop virtual channel dependency according to the present invention.
Fig. 8B is a schematic diagram of a deadlock configuration on a loop before adding C _ s to a new loop virtual channel of a fast path according to the present invention.
Fig. 8C is a schematic diagram of the virtual channels on the loop before the new loop virtual channel of the fast path is added with C _ s according to the present invention.
Fig. 8D is a schematic diagram of the dependency relationship of the virtual channels on the loop before the new loop virtual channel of the fast path adds C _ s according to the present invention.
Fig. 8E is a schematic diagram of the virtual channels on the loop after the new loop virtual channel of the fast path is added with C _ s according to the present invention.
Fig. 8F is a schematic diagram of the dependency relationship of the virtual channels on the loop after the fast path new loop virtual channel C _ s according to the present invention.
Fig. 9 is a flowchart of an embodiment of a virtual channel switching method.
Fig. 10 is a schematic diagram illustrating an example of a high-dimensional direct network routing forwarding implementation according to the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more clearly understood, the following describes in detail a method and a system for adaptive interconnection and route control for direct network fast path link resources according to the present invention with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The high-dimensional switch is an important unit for building a high-dimensional direct network. And building a link, namely a fast path, for directly connecting originally non-adjacent nodes in the direct network in the high-dimensional direct network. The invention provides an interconnection method among high-dimensional switches and a corresponding deadlock-free routing method.
Specifically, the invention firstly carries out grouping interconnection on nodes in each high-dimensional switch in the high-dimensional direct network; then, information of the node pairs connected through the fast path is stored in the mapping table. And finally, carrying out routing selection according to the grouping result, the information provided by the mapping table and the fixed sequence of the dimensionality, calculating output ports and virtual channels to other nodes for each node in the network, and realizing deadlock-free routing forwarding. As shown in fig. 1, the present invention includes three sub-modules: the system comprises a high-dimensional switch interconnection module, a routing information generation module and a routing table generation module.
And the high-dimensional switch interconnection module groups the nodes in the high-dimensional switch according to parameter configuration and sets a communication node in each group. It has the characteristics of three: firstly, the number of nodes in each group is the same; second, there is no intersection of paths from group to group; and thirdly, the Euler distance between the communication node and the group of nodes is minimum. This grouping balances the group load and minimizes the packet hop count. Grouping information is provided for the route forwarding table generation method by grouping nodes in the high-dimensional switch. The grouping information includes coordinate values of all nodes including the group communication node. And the given node can find the group in which the given node is located according to the grouping information and the group communication node.
And the routing information generation module records the nodes connected by the fast path in a mapping table. For convenience of description, nodes connected through a fast path are defined as sibling nodes. The brother node information recorded by the mapping table stores the connection relation on the fast path and provides necessary information for the method for generating the routing forwarding table.
The routing table generating module generates a routing forwarding table and a virtual channel switching table for each node by using the grouping information provided by the high-dimensional switch interconnection module and the mapping table provided by the routing information generating module.
The present invention provides a method for averagely dividing nodes in each high-dimensional switch into a plurality of small groups for a high-dimensional switch interconnection module. In the method, an input parameter N is used, which represents the group number in each high-dimensional switch and is also the link number of the fast path in the high-dimensional switch. The method can adaptively perform fast link interconnection and grouping according to the value of N. In summary, the grouping method selects a communication node for each group, and then sequentially adds the closest node to the group based on the communication node. The specific implementation details are as shown in fig. 2, the grouping result is shown in fig. 3, fig. 4 shows the grouping process when N is 4, and the detailed steps are as follows:
step 101: reading input parameters: the input parameter N is read. Taking the example (d) of fig. 3 as an example, the input parameter N is 4.
Step 102: selecting a starting point: and selecting any point in the high-dimensional switch, adding the point into the current group Gi, and using the point as a communication node. In the example of fig. 3 (d), the node with coordinates (0, 0, 0) in the high-dimensional switch is selected to be added to the first group G0 and used as its correspondent node.
Step 103: it is determined whether the current team is the last team in the high dimensional switch. According to the number of nodes to be distributed in each group, the number of groups to which the rest nodes need to be distributed can be calculated, and whether the current group is the last group or not is judged. In the example of fig. 3 (d), each subgroup should be assigned 3 nodes, and there are 11 unassigned nodes, so the current subgroup is not the last group.
Step 104: all ungrouped nodes are added into the current group: and if the current group is the next group, adding all the remaining nodes into the current group.
Step 105: the farthest point from the Gi communication node is selected, added to Gi +1, and used as its communication node. In the example of fig. 3 (d), node (1, 2, 1) is selected, added to the second subgroup G1, and acts as its correspondent node.
Step 106: and adding nodes to Gi and Gi +1 in sequence according to the position and the dimension sequence of the communication node. And selecting the node closest to the Gi communication node from the ungrouped nodes to join the group Gi, and selecting one of the nodes according to the dimension sequence when the communication node is closest to the plurality of nodes at the same time. The same operation was then performed on the Gi +1 subgroup. In the example of fig. 3, the nodes are selected from the nodes closest to the corresponding node in the order of the dimensions a, c, b, and the corresponding packets are added. When two nodes in the b-dimension exist at the same time, a node having the same b-dimension coordinate value as that of other packet communication nodes is preferentially selected. In the example of (d) in fig. 3, (1, 0, 0) node is selected to join the group G0, and (0, 2, 1) node is selected to join the group G1.
Step 107: and judging whether the current grouping is full or not and whether the number of the nodes in the current grouping reaches the number of the nodes to be distributed in each group or not.
Step 108: and judging the grouping end, and judging whether all the nodes in the high-dimensional switch are grouped.
Step 109: and traversing the grouping, and selecting a new group of communication nodes in the ungrouped nodes according to the dimension sequence and the positions of the existing communication nodes. And selecting the node closest to the previous group of communication nodes by the new group of communication nodes, and selecting one of the nodes according to the dimensional sequence when a plurality of nodes meet the requirements. In the example of fig. 3, a new correspondent node is selected from nodes closest to the last group of correspondent nodes in the order of a, c, and b dimensions, and when there are nodes in both b dimensions, a node having a different b coordinate value from the other group of correspondent nodes is preferentially selected.
And step 110, whether the groups are crossed or not is checked, and whether the nodes in each group can form a connected graph or not is checked. In the example of fig. 4 (b), there is an intersection between packets.
Step 111: and adjusting the dimension sequence. And d, adjusting the dimension sequence to be b, a and c.
For a routing information generation module, the present invention provides a method for recording node information connected through a fast link, the specific implementation details are as shown in fig. 5, for example, as shown in fig. 6, and the detailed steps are as follows:
step 201: swi traverses the switching node. An initial node is selected and the switching nodes are traversed in the network. In the example of fig. 6, the (0, 0, 0, 1, 0) node is selected as the initial node.
Step 202: swj traverse the switching nodes. An initial node is selected and the switching nodes are traversed in the network. In the example of fig. 6, the (0, 0, 0, 4, 1) node is selected as the initial node.
Step 203: determine whether Swi and Swj are sibling nodes: it is checked whether Swi and Swj are nodes at both ends of the same fast link. In the example of fig. 6, (0, 0, 0, 1, 0) and (0, 0, 0, 4, 1) are sibling nodes.
Step 204: mapping table record { Swi, Swj }: if Swi and Swj are siblings, { Swi, Swj } is recorded in the mapping table. In the example of fig. 6, one record { (0, 0, 0, 1, 0), (0, 0, 0, 4, 1) } is added in the mapping table.
Step 205: swi traverses the end judgment.
Step 206: swj go through the end decision.
The invention provides a method for generating a routing forwarding table, which aims at a routing table generating module. The main characteristics are three: (1) and when the packet where the destination node is located is not connected with the fast link, generating a routing forwarding table according to a traditional dimension order routing method. (2) When the packet where the destination node is located has a fast link connection, three-stage routing is required: the first stage, from the source node to one end of the fast link, the other end of the fast link must be the communication node of the group where the destination node is located; in the second stage, the communication nodes of the group where the destination node is located are reached from one end of the fast link to the other end of the fast link, namely through the fast link; and the third stage, from the communication node of the group where the destination node is located to the destination node. (3) And when the hop count of the three-stage route is greater than the hop count of the traditional dimension order route, generating a route forwarding table according to the traditional dimension order route, otherwise, generating the route forwarding table according to the three-stage route mode.
In summary, the routing method first determines whether a group of a destination node has a fast path, if no fast path exists, a forwarding table is generated by using a conventional dimension order route, and if a fast path exists, the path length of a three-stage route and the path length of a conventional dimension order route are compared. When the route of the three-stage route is short, judging which stage of the three stages of route forwarding of the current node is, performing route calculation of the corresponding stage, and generating a forwarding table entry.
The specific implementation details are shown in fig. 7, and the detailed steps are as follows:
step 301: the current node Swi, the destination node Swj are input.
Step 302: the communication node SwCom is acquired Swj. And acquiring the communication node SwCom of the group where the destination node is positioned by utilizing the grouping information provided by the high-dimensional switch interconnection module.
Step 303: it is determined whether SwCom has a sibling node. And judging whether the SwCom has nodes connected through the fast link or not according to a mapping table provided by the routing information generation module.
Step 304: the three-stage route Hop count Hop1 is calculated. The three-stage route consists of three parts, namely the Hop count from Swi to SwCom brother node, the Hop count from SwCom brother node to SwCom, the Hop count from SwCom to Swj and Hop1 which is the sum of the Hop counts of three sections.
Step 305: and calculating the Hop step number Hop2 of the traditional dimension order route.
Step 306: hop1 and Hop2 sizes were compared. If Hop1 is greater than Hop2, this indicates that the current node is not in the three-phase route, or that the current node is in the third phase of the three-phase route.
Step 307: it is determined whether Swi is a SwCom sibling node. If Swi is SwCom brother node, it indicates that the current node is in the second stage of the three-stage route, otherwise, it indicates that the current node is in the first stage of the three-stage route.
Step 308: and the second stage routing method calculates a routing forwarding table entry. The current node is in the second phase of the three-phase route, and the output port is a fast link port.
Step 309: the first stage routing method calculates the route forwarding table entry. The current node is in the first phase of the three-phase route, selecting output ports from the current node to the SwCom sibling node in order of dimensionality.
Step 310: the traditional dimension order routing method calculates a routing forwarding table entry.
The invention provides a virtual channel switching method aiming at a routing table generation module. The virtual channels are realized by a pair of message buffers on the switching nodes at both ends of the link, and a plurality of virtual channels can share one link. In summary, to avoid routing deadlocks, the method uses three virtual lanes C _ dor0, C _ dor1, and C _ s. In the three-stage routing, the first stage uses the virtual channels C _ dor0 and C _ dor1 in the traditional dimension order routing, and the second and third stages use the virtual channel C _ s. As shown in fig. 8, the virtual channel C _ s solves the problem of routing deadlock caused by a new loop introduced by the fast path.
The specific implementation details are as shown in fig. 9, and the detailed steps are as follows:
step 401: the known input port, output port, current virtual channel C _ now.
Step 402: and judging whether the output port is a fast link port or not.
Step 403: and outputting the C _ s. If the output port is a fast link port, the current node is in the second stage of the three-stage routing, and the virtual channel is switched to the C _ s.
Step 404: whether C _ now! C _ s and switch dimensions. And judging whether the current virtual channels C _ now and C _ s are different and whether the input port and the output port belong to different dimensions.
Step 405: output C _ dor 0. If the condition C _ now! C _ s and switch dimension, indicating that the current node is not in the three-phase route, or in the first phase of the three-phase route, the virtual channel is switched to C _ dor 0.
Step 406: whether dataline is crossed. And judging whether the link where the output port is located crosses dataline.
Step 407: output C _ dor 1. If dataline is crossed, the virtual channel is switched to C _ dor 1.
Step 408: and outputting C _ now. The current virtual channel is kept unchanged.
Fig. 10 is an example of the generation of a forwarding path between a source node and a destination node using the routing algorithm of the present invention. In the example, it can be seen that the routing algorithm has the following characteristics:
only when the communication node of the group where the destination node is located has a fast path and the path length from the fast path to the destination node is not greater than the path length provided by the traditional dimension order routing method, the fast path is used for routing forwarding, otherwise, the routing forwarding is carried out according to the traditional dimension order routing method.
Not all point-to-point communications are shortest paths using fast paths. As in fig. 9, the path from the source node 2 to the destination node 2 is not the shortest path in the network.
The invention can achieve the following beneficial effects:
the high-dimensional switch interconnection method has strong adaptability, and can provide a flexible rapid path networking method according to the user requirements and the condition of link resources; the routing algorithm has strong compatibility and is suitable for the interconnection scheme of high-dimensional switches under the condition of aiming at different link resources; the routing algorithm is free of deadlock, and deadlock-free characteristics are still guaranteed under the condition that the fast path introduces more loops.
Claims (6)
1. A method for controlling adaptive interconnection and routing is characterized by comprising the following steps:
in the interconnection step of the high-dimensional switch, a plurality of nodes of the high-dimensional switch are selected as communication nodes, each communication node is taken as a grouping identifier, all the nodes of the high-dimensional switch are divided into a plurality of groups, and the following conditions are met: each group has the same number of nodes, the paths between the nodes of the group are not crossed with the paths between the nodes of other groups, and the Euler distance between the nodes of the group and the communication node of the group is smaller than the Euler distance between the nodes of the group and the communication node of other groups;
a step of generating routing information, in which any node in the communication nodes is taken as a node A, and the node A traverses all the communication nodes; taking any node except the node A in the communication nodes as a node B, and traversing all the communication nodes by the node B; if the node A and the node B are nodes at two ends of the same direct connection link, the node A and the node B are brother nodes, and the direct connection link is a fast path between the node A and the node B; obtaining a quick path between any pair of brother nodes, and generating a brother node information mapping table by all the quick paths;
a step of generating a routing table, which is to establish a routing forwarding table and a virtual channel switching table of the high-dimensional switch by using the grouping information and the sibling node information mapping table; the method comprises the steps that a communication node of a group where a routing path starting node is located is used as a source node, a communication node of a group where a routing path terminal node is located is used as a destination node, whether a fast path exists between the source node and the destination node or not is judged, when only a traditional dimensional sequence routing path exists or the fast path exists and the number of steps of the fast path is larger than that of the traditional dimensional sequence routing path, routing is carried out through the traditional dimensional sequence routing path, and otherwise, the fast path is selected for routing;
and a route forwarding step, judging whether the route path forwarded by the information has the fast path or not according to the route forwarding table and the virtual channel switching table, so as to use the fast path or the traditional dimension order route path for routing.
2. The adaptive interconnect and route control method according to claim 1, wherein the high-dimensional switch interconnect step specifically includes: setting a grouping parameter N, selecting N nodes with evenly distributed positions in the high-dimensional switch as N communication nodes, and averagely dividing all the nodes into N groups by taking each communication node as a grouping identifier; wherein N is a positive integer.
3. The adaptive interconnect and route control method of claim 1, wherein the route forwarding step further comprises: a virtual channel switching step, dividing the fast path into three-stage routes, including a first stage from the initial node to the source node; a middle segment from the source node to the destination node; a terminal node from the destination node to the terminal node; and establishing a first virtual channel, a second virtual channel and a third virtual channel, wherein the first virtual channel and the second virtual channel are used in the initial segment, and the third virtual channel is used in the middle segment and the final segment.
4. An adaptive interconnection and routing control system, comprising:
the high-dimensional switch interconnection module is used for selecting a plurality of nodes of the high-dimensional switch as communication nodes, taking each communication node as a grouping identifier, dividing all the nodes of the high-dimensional switch into a plurality of groups, and meeting the following conditions: each group has the same number of nodes, the paths between the nodes of the group are not crossed with the paths between the nodes of other groups, and the Euler distance between the nodes of the group and the communication node of the group is smaller than the Euler distance between the nodes of the group and the communication node of other groups;
the routing information generation module is used for establishing a brother node information mapping table of the high-dimensional switch, taking any one of the communication nodes as a node A, and traversing all the communication nodes by the node A; taking any node except the node A in the communication nodes as a node B, and traversing all the communication nodes by the node B; if the node A and the node B are nodes at two ends of the same direct connection link, the node A and the node B are brother nodes, and the direct connection link is a fast path between the node A and the node B; obtaining a quick path between any pair of brother nodes, and generating the brother node information mapping table by all the quick paths;
a routing table generating module, configured to establish a routing forwarding table and a virtual channel switching table of the high-dimensional switch according to the packet information and the sibling node information mapping table; the method comprises the steps that a communication node of a group where a routing path starting node is located is used as a source node, a communication node of a group where a routing path terminal node is located is used as a destination node, whether a fast path exists between the source node and the destination node or not is judged, when only a traditional dimensional sequence routing path exists or the fast path exists and the number of steps of the fast path is larger than that of the traditional dimensional sequence routing path, routing is carried out through the traditional dimensional sequence routing path, and otherwise, the fast path is selected for routing;
and the route forwarding module is used for judging whether the route path forwarded by the information has the fast path or not according to the route forwarding table and the virtual channel switching table so as to use the fast path or the traditional dimension order route path for routing.
5. The adaptive interconnection and routing control system according to claim 4, wherein the high-dimensional switch interconnection module specifically comprises: setting a grouping parameter N of the high-dimensional switch, selecting N nodes with evenly distributed positions as N communication nodes, and evenly dividing all the nodes of the high-dimensional switch into N groups by taking each communication node as a center; wherein N is a positive integer.
6. The adaptive interconnect and route control system of claim 4, wherein the route forwarding module further comprises:
the virtual channel switching module is used for constructing a plurality of virtual channels to avoid routing deadlock; dividing the fast path into three-stage route including initial segment from the initial node to the source node; a middle segment from the source node to the destination node; a terminal node from the destination node to the terminal node; and establishing a first virtual channel, a second virtual channel and a third virtual channel, wherein the first virtual channel and the second virtual channel are used in the initial segment, and the third virtual channel is used in the middle segment and the final segment.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711320765.3A CN109842553B (en) | 2017-12-12 | 2017-12-12 | Self-adaptive interconnection and routing control method and system for link resources |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711320765.3A CN109842553B (en) | 2017-12-12 | 2017-12-12 | Self-adaptive interconnection and routing control method and system for link resources |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109842553A CN109842553A (en) | 2019-06-04 |
| CN109842553B true CN109842553B (en) | 2021-10-08 |
Family
ID=66882876
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711320765.3A Active CN109842553B (en) | 2017-12-12 | 2017-12-12 | Self-adaptive interconnection and routing control method and system for link resources |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109842553B (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103124420A (en) * | 2013-01-21 | 2013-05-29 | 电子科技大学 | Wireless on-chip network structuring method |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8139490B2 (en) * | 2009-12-21 | 2012-03-20 | Google Inc. | Deadlock prevention in direct networks of arbitrary topology |
| CN104079491B (en) * | 2014-07-07 | 2018-04-27 | 中国科学院计算技术研究所 | A kind of router and method for routing towards high-dimensional network |
| CN104539536B (en) * | 2014-12-01 | 2017-10-17 | 清华大学 | The stream control of dynamical state driving and Torus network self-adapting routing methods |
| CN107094116B (en) * | 2017-05-26 | 2020-02-28 | 中国科学院计算技术研究所 | Direct network routing method and system containing cross-dimension link |
-
2017
- 2017-12-12 CN CN201711320765.3A patent/CN109842553B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103124420A (en) * | 2013-01-21 | 2013-05-29 | 电子科技大学 | Wireless on-chip network structuring method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109842553A (en) | 2019-06-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6267367B2 (en) | Packet routing method in distributed direct interconnection network | |
| US9825844B2 (en) | Network topology of hierarchical ring with recursive shortcuts | |
| EP1579716B1 (en) | Routing scheme based on virtual space representation | |
| Sancho et al. | A new methodology to compute deadlock-free routing tables for irregular networks | |
| KR101809779B1 (en) | Automated traffic engineering for 802.1aq based upon the use of link utilization as feedback into the tie-breaking mechanism | |
| CN111147372B (en) | Downlink message sending and forwarding method and device | |
| CN108111410B (en) | Method and device for constructing deadlock-free route in network with Cartesian topology | |
| Palesi et al. | A method for router table compression for application specific routing in mesh topology NoC architectures | |
| KR20150030644A (en) | Tie-breaking in shortest path determination | |
| US9529775B2 (en) | Network topology of hierarchical ring with gray code and binary code | |
| Nosrati et al. | G-CARA: A Global Congestion-Aware Routing Algorithm for traffic management in 3D networks-on-chip | |
| US10635774B2 (en) | Integrated circuit design | |
| US9529774B2 (en) | Network topology of hierarchical ring with gray coding shortcuts | |
| CN112637061B (en) | Dynamic multi-factor path calculation method based on heuristic algorithm | |
| TWI661700B (en) | Network topology system and topology building method thereof | |
| CN107094116B (en) | Direct network routing method and system containing cross-dimension link | |
| CN109842553B (en) | Self-adaptive interconnection and routing control method and system for link resources | |
| US10855581B2 (en) | System and method of computing ethernet routing paths | |
| Gerla et al. | Routing in the bidirectional shufflenet | |
| CN100518382C (en) | Shortest path searching method and device under multi-restraint conditions in automatic switching optical network | |
| KR102001487B1 (en) | Method for controlling software defined networking and computing device performing the same | |
| US11765103B2 (en) | Large-scale network with high port utilization | |
| US20230068443A1 (en) | Dynamic packet routing using prioritized groups | |
| US20240154903A1 (en) | Cyclic dragonfly and megafly | |
| CN113824781A (en) | Data center network source routing method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |