WO2017128112A1

WO2017128112A1 - Calculation method, system, device, and chip for combined programming action in software-defined network

Info

Publication number: WO2017128112A1
Application number: PCT/CN2016/072353
Authority: WO
Inventors: 潘恒; 贺鹏; 关洪涛; 李振宇; 谢高岗
Original assignee: 中国科学院计算技术研究所
Priority date: 2016-01-27
Filing date: 2016-01-27
Publication date: 2017-08-03

Abstract

Provided are a calculation method, system, device, and chip for combined programming action in a software-defined network, relating to the technical field of software-defined networking. The method comprises: abstracting a rule-action linked list in a software-defined network, and generating one or more nodes, wherein the nodes form a node set V; and adding directed edges to all nodes in the node set V to generate directed graphs, and generating Hamiltonian paths for the directed graphs, wherein a sum of weights of all edges in the directed graphs is a minimum. The method ensures, by means of a series of theoretical models, the semantic equivalence of an action list of a combination rule in SDN combined programming, and one Hamiltonian path is searched for in an abstracted directed graph to calculate an action list of a final combination rule, and therefore, the action list can guarantee that the number of actions of the combination rule are at a minimum.

Description

Software defined network combined programming action calculation method, system, device and chip

Technical field

The present invention relates to the field of software-defined network technologies, and in particular, to a software-defined network combined programming action calculation method, system, device and chip.

Background technique

With the continuous development of information technology, the Internet has become an indispensable information infrastructure for modern society. However, the current network thin waist architecture has been unable to carry more and more network needs of users, therefore, a new type of network architecture - Software-defined networking (SDN) has received extensive attention. The main idea of SDN is to separate the control logic in traditional network devices from the data plane and manage the entire network through centralized control.

Flexible programmability is one of the important features provided by SDN. Therefore, the SDN network programming model has become a hot research field. Among them, modular combination programming has become the most important programming feature in the network programming model. In modular combination programming, it is mainly divided into parallel programming and serial programming. Parallel programming is mainly to realize that multiple modules process the same data packet in parallel according to their respective logics, while serial processing refers to a data packet processed by one module logic and then processed by the next module. In modular combination programming, parallel The SDN switch rules generated by the module and the serial module need to be compiled into a logically equivalent set of rules and sent to the underlying switch. The existing rule compiling algorithm mainly generates new ones according to the intersection of the rule table spaces of multiple submodules. Rule set, specifically:

(1) When the parallel module is compiled, each module first generates its own rule table. Then, each two modules are forked according to their respective rule tables, that is, the matching domain space of each two rules from the two rule tables is intersected. If the result of the intersection is not empty, a new rule is generated according to the intersection of the two rules, as shown in Figure 1-1.

(2) When the serial module is compiled, each module still generates its own rule table first, and then preprocesses the rule table corresponding to the module that needs to process the data packet first, that is, each rule of the rule table is used by the rule table. The corresponding action first acts on the matching field of the rule. Finally, the rule table of the two modules is multiplied. The process of forming the rule table by fork is similar to the compilation of the parallel module, as shown in Figure 1-2.

The existing SDN combination programming compilation algorithm is based on the intersection of rule spaces of different submodules. Constructing the merged rule and calculating the priority size for the merged rule. However, the current compilation algorithm calculates the action list of the merged rule only for the action rules of the two rules of the submodule. Simply concatenate together, as shown in Figure 2-1.

This simple method of concatenating the action list will cause the synthesized rule to be not equivalent to the semantics of the previous rule or to generate redundant actions. The most fundamental reason for this problem is that multiple parallel modules may need a packet header for the packet. Simultaneous reading and writing and then forwarding to different ports, and this method of serially arranging the action list makes the two subaction lists unable to form independent operations on the data packets. Therefore, the logical parallel operation cannot be guaranteed at all. As shown in 2-2, the requirement of rule 1 is to modify the F ₂ matching field of the original data packet and then forward it to port 1, and rule 2 is to forward the original data packet to its matching domain F _{1 and} then forward it to port 2. However, the data packet forwarded by the merged rule to port 2 is modified by F ₁ and F ₂ at the same time, which is not equivalent to the semantics of the original rule 2.

In addition, although this transfer method can guarantee the modular compilation of the serial mode, multiple sub-action lists may repeatedly operate on the packet header, thereby generating redundant actions, as shown in Figure 2-3. The first modification to the packet F ₁ in the action list of the merge rule is redundant.

Invention disclosure

In view of the deficiencies of the prior art, the present invention proposes a software defined network combination programming action calculation method, system, device and chip.

The invention provides a software definition network combination programming action calculation method, which comprises:

Step 1, abstracting the rule action list in the software-defined network, generating one or more nodes, the nodes forming a node set V;

Step 2: adding a directed edge to all the nodes in the node set V, generating a directed graph, and generating a Hamilton path for the directed graph, wherein a sum of weights of each edge in the directed graph is minimum .

The step 1 includes the step 101 of acquiring the original data p, and generating an initial node n ₀ for the original data p, wherein the data packet header of the original data p is copied and associated with the initial node n ₀ . Executing the rule action list in sequence;

Step 102, if the rule in the rule action list is empty, the operation is ended, otherwise step 103 is performed;

Step 103: Record the action to be performed by the rule action list as act. If the act is modify, modify the packet header of the original data p. If the act is forward, generate a new node n. _i , and copying the packet header of the original data p, and associating the original data p with the node n _i , and jumping to step 102 until the rule in the rule action list is empty.

Step 2 includes step 201, if the node set V is empty or only 1 node, the operation ends, otherwise step 202 is performed;

Step 202, taking one node v from the node set V, and sequentially performing the following operations on the node v and the remaining nodes in the set, wherein the remaining nodes are denoted as v _i : if the node v is associated The packet header of the original data p is converted into the packet header of the original data p associated with v _i by adding a modify action, and a directed edge is added, and the node v points to the node v. _i , and vice versa;

In step 203, the weight of the added directed edge is calculated, and the value of the weight is the data of the added modifyaction, and the process proceeds to step 201.

Each of the nodes is a <packet, port> pair.

The invention also provides a software defined network combined programming action computing system, comprising:

Generating a node set V module for abstracting a rule action linked list in the software defined network to generate one or more nodes, the nodes forming a node set V;

Generating a Hamilton path module for adding a directed edge to all of the nodes in the set of nodes V, generating a directed graph, generating a Hamilton path for the directed graph, wherein each edge of the directed graph The sum of the weights is the smallest.

The generating node set V module includes an associated n ₀ module for acquiring the original data p, and generating an initial node n ₀ for the original data p, wherein the data packet header of the original data p is copied, and the The initial node n ₀ is associated, and the rule action linked list is sequentially executed;

a judging module, configured to end the operation if the rule in the rule action list is empty, otherwise execute the associated n _i module;

The associated n _i module is configured to record an action to be performed by the rule action list as act, and if the act is modify, modify a packet header of the original data p, and if the act is forward, generate a a new node n _i , and copying the packet header of the original data p, and associating the original data p with the node n _i , jumping to the judging module until the rule action list The rule is empty.

The generating a Hamilton path module includes a determining a node number module, configured to end the operation if the node set V is empty or has only one node, and otherwise execute the added directed edge module;

Adding a directed edge module for taking one node v from the node set V, and sequentially performing the following operations on the node v and the remaining nodes in the set, wherein the remaining nodes are recorded as v _i : The packet header of the original data p associated with the node v is converted into a packet header of the original data p associated with v _i by adding a modify action, and a directed edge is added, by the node v Point to the node v _i and vice versa;

The calculation weight module is configured to calculate the weight of the added directed edge, and the value of the weight is the data of the added modify action, and the parameter is determined by the node number module.

Each of the nodes is a <packet, port> pair.

The present invention also provides an apparatus for a software defined network combined programming action computing system.

The invention also proposes a chip for software-defined network combination programming action calculation method.

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1-1 is a flowchart of parallel module compilation;

Figure 1-2 shows the serial module compilation flowchart;

Figure 2-1 is a flow chart of compiling action list;

Figure 2-2 is a semantic inequality diagram;

Figure 2-3 is a redundant view of the action;

Figure 3 is an abstract view of the rule action list;

Figure 4 is a construction diagram of a directed graph;

Figure 5 is a Hamilton road map;

Figure 6 is a flow chart of the present invention;

Figure 7 is an illustration of an example of the invention.

The best way to implement the invention

The present invention abstracts the action list to produce multiple nodes (each node is a <packet, port> pair). Suppose a raw data p hits the rule r. After the hit, the action list associated with r is executed, denoted as r.a. The specific abstraction process is as follows:

Step 1: Corresponding to the original data p, the initial node n _{0 is} generated, that is, the data packet header of p is copied, and then associated with n ₀ , and then ra is sequentially executed;

Step 2: If the rule in r.a is empty, jump to step four, otherwise perform step three;

Step 3: Remember that the action to be executed by ra is act. If act is modify (that is, modify the packet instruction), modify the packet header of packet p. If act is forward (that is, forward the packet instruction), a new one is generated. Node n _i , and the packet header of packet p is copied and associated with node n _i . Jump to step two;

Step 4: End.

The whole process is shown in Figure 3.

After the multiple action lists are subjected to the above steps, one or more nodes can be generated, and these nodes can form a node set V. Then, add a directional edge between these nodes, and finally form a directed graph, the specific steps are as follows.

Step 1: If the set V is empty or has only 1 node, jump to step 4, otherwise proceed to step 2;

Step 2: Take 1 node v from the set V, and then v and the remaining nodes in the set (set to v _i ) are sequentially performed as follows: if the packet header of the v association can be added by adding a modify action (ie, modifying the data) The way of the packet header instruction is changed to the packet header associated with v _i , then a directed edge is added, and v points to vi; vice versa. Go to step three;

Step 3: Calculate the weight of the added directed edge, the size is the data of the added modify action, and go to step one;

Step 4: End

The entire process is shown in Figure 4.

After the above steps, one directed graph can be constructed. Then, starting from the node n0 associated with the original data packet p, a Hamiltonian path is sought, which traverses all the nodes in the graph, and the sum of the weights of the sides is the smallest, as shown in FIG.

By finding a Hamiltonian path, you can refactor a combined action list. The action list is mainly composed of the modify action on the side of the path and the derived from each node. The forward action (that is, forwarding packet instructions) consists of the order of the paths.

In summary, the entire flow chart is shown in Figure 6.

A specific matter is shown in Fig. 7, for example.

Each of the nodes is a <packet, port> pair.

The present invention also proposes an apparatus comprising a software defined network combined programming action computing system.

The invention also proposes a chip using a software defined network combination programming action calculation method.

Industrial applicability

The software-defined network combined programming action calculation method, system, device and chip proposed by the invention have the following advantages and applicability:

Through a series of theoretical modeling, the present invention can ensure the semantic equivalence of the synthetic rule action list in the SDN combination programming, and calculate the action list of the final synthesis rule by searching for a Hamiltonian path in the abstract directed graph. The action list can guarantee the minimum number of actions.

Claims

A software-defined network combined programming action calculation method, comprising:

Step 1, abstracting the rule action list in the software-defined network, generating one or more nodes, the nodes forming a node set V;

Step 2: adding a directed edge to all the nodes in the node set V, generating a directed graph, and generating a Hamilton path for the directed graph, wherein a sum of weights of each edge in the directed graph is minimum .
The software-defined network combination programming action calculation method according to claim 1, wherein the step 1 comprises the step 101 of acquiring raw data p, and generating an initial node n 0 for the original data p, wherein the original The packet header of the data p is copied and associated with the initial node n 0 , and the rule action list is sequentially executed;

Step 102, if the rule in the rule action list is empty, the operation is ended, otherwise step 103 is performed;

Step 103: Record the action to be performed by the rule action list as act. If the act is modify, modify the packet header of the original data p. If the act is forward, generate a new node n. i , and copying the packet header of the original data p, and associating the original data p with the node n i , and jumping to step 102 until the rule in the rule action list is empty.
The software-defined network combination programming action calculation method according to claim 1, wherein the step 2 includes a step 201, if the node set V is empty or has only one node, the operation ends, otherwise step 202 is performed. ;

Step 202, taking one node v from the node set V, and sequentially performing the following operations on the node v and the remaining nodes in the set, wherein the remaining nodes are denoted as v i : if the node v is associated The packet header of the original data p is converted into the packet header of the original data p associated with v i by adding a modify action, and a directed edge is added, and the node v points to the node v. i , and vice versa;

In step 203, the weight of the added directed edge is calculated, and the value of the weight is the data of the added modifyaction, and the process proceeds to step 201.
The software-defined network combination programming action calculation method according to claim 1, wherein each of said nodes is a <packet, port> pair.
A software defined network combined programming action computing system, comprising:

Generating a node set V module for abstracting a rule action linked list in the software defined network to generate one or more nodes, the nodes forming a node set V;

Generating a Hamilton path module for adding a directed edge to all of the nodes in the set of nodes V, generating a directed graph, generating a Hamilton path for the directed graph, wherein each edge of the directed graph The sum of the weights is the smallest.
The software-defined network combined programming action computing system of claim 5, wherein the generating node set V module comprises an associated n 0 module for acquiring raw data p, and generating an initial node n for the original data p 0 , wherein a packet header of the original data p is copied and associated with the initial node n 0 , and the rule action linked list is sequentially executed;

a judging module, configured to end the operation if the rule in the rule action list is empty, otherwise execute the associated n i module;

The associated n i module is configured to record an action to be performed by the rule action list as act, and if the act is modify, modify a packet header of the original data p, and if the act is forward, generate a a new node n i , and copying the packet header of the original data p, and associating the original data p with the node n i , jumping to the judging module until the rule action list The rule is empty.
The software-defined network combination programming action calculation system according to claim 5, wherein said generating a Hamilton path module comprises determining a node number module for ending if said node set V is empty or has only one node Operation, otherwise perform the addition of the directed edge module;

Adding a directed edge module for taking one node v from the node set V, and sequentially performing the following operations on the node v and the remaining nodes in the set, wherein the remaining nodes are recorded as v i : The packet header of the original data p associated with the node v is converted into a packet header of the original data p associated with v i by adding a modify action, and a directed edge is added, by the node v Point to the node v i and vice versa;

The calculation weight module is configured to calculate the weight of the added directed edge, and the value of the weight is the data of the added modify action, and the parameter is determined by the node number module.
The software-defined network combined programming action computing system of claim 5 wherein each of said nodes is a <packet, port> pair.
An apparatus comprising a system according to any of claims 5-8.
A chip using the method of any of claims 1-4.