CN100456840C - Method to split LPM algorithm to two CPUs - Google Patents

Abstract

The invention provides a method for splitting an LPM algorithm into two CPUs, which belongs to the field of computer network communication. Maintain the parent-child relationship of the routing prefix. Before updating the routing table entry, first look up the FIB table on the control CPU to determine whether there is a prefix shorter than the current routing prefix, that is, the parent prefix. If it exists, set the mask length of the parent prefix to , and the index corresponding to the routing table entry are put into the distributed data structure, and the network processor is notified together to implement the LPM algorithm. The longest match is achieved through the coordination of the two CPUs. On the premise of ensuring that the routing table is completely correct, the forwarding capability of the NP is maximized, and the work that is not related to the longest match forwarding process is eliminated from the NP.

Description

Method to split LPM algorithm to two CPUs

Technical field

The invention belongs to the field of computer network communication, and in particular relates to a method for splitting an LPM algorithm into two CPUs.

Background technique

In the IP switching network, the network layer forwarding of the IP data packet is completed by searching the routing table in the ASIC chip. In order to expand the routing scale, a network processor (hereinafter referred to as NP) is usually needed. NP can accommodate at least 100,000 routing table items. In order to improve the search speed, these routing table items are stored in the chip memory in a multi-level search tree to implement LPM (Longest Prefix Match). The multi-level search tree needs to be assisted by another data structure to record the parent-child relationship of routing prefixes. Undoubtedly, this work reduces the ability of NP to process data, because the parent-child relationship structure is completely invisible in the forwarding process, and at the same time, every increase in NP A route needs to find the parent-child structure, and record the routing information on the corresponding node; and every time a route NP is deleted, it also needs to find the parent-child structure, take out the parent node information and re-modify the multi-level search tree. This will greatly reduce the forwarding efficiency of the CPU during route flapping, because when routing table entries are added, deleted or modified, routing table entries cannot be queried and data packets forwarded, that is, any overhead that affects NP routing table entry lookups will become a bottleneck in the forwarding process.

Contents of the invention

Purpose of the present invention: to provide a method for splitting the LPM algorithm into two CPUs, to ensure that the routing table is completely correct, to maximize the forwarding capability of the NP, and to remove the work that has nothing to do with the longest matching forwarding process from the NP remove.

Technical content of the present invention: a kind of method that LPM algorithm is split into two CPUs, comprising:

(1) two CPUs comprise network processor and control CPU, guide the control CPU that network processor routing table is set to complete the work of maintaining the parent-child relationship of routing prefix;

(2) Before updating the routing table entry, first search the structure on the control CPU to determine whether there is a prefix shorter than the current routing prefix, that is, the father prefix. If it exists, the mask length of the father prefix and the corresponding routing table The index of the item is put into the distributed data structure, and the network processor is notified together to realize the LPM algorithm.

The network processor forwarding table index is stored in the FIB table of the control CPU.

Further includes:

(1) When adding a route, the control CPU first searches the FIB table. If the routing prefix already exists, it takes out the network processor forwarding table index stored in the FIB table, and notifies the network processor together with the routing table entry, and the network processor enters the update Process, if the routing prefix does not exist in the FIB table of the controlling CPU, the network processor is notified to enter the adding process, and the network processor uses the LPM algorithm to create a new forwarding entry, and write the entry index into the multi-level search tree In the node, the routing table entry is quickly located in case of routing lookup, and the index is returned to the control CPU at the same time, and the control CPU writes the network processor forwarding table index together with the entire routing table entry into its own FIB table.

(2) When deleting a route, the control CPU first searches the FIB table to see if there is a shorter routing prefix, and if it exists, it takes out the father prefix information stored therein and informs the network processor, and the network processor indexes the father prefix forwarding table according to the LPM algorithm Revert to the search tree node.

Technical effect of the present invention: the work of maintaining the parent-child relationship is handed over to the control CPU to complete, before updating the routing table entry, first search for the structure on the control CPU, if there is a father prefix, the mask length of the father prefix and the corresponding routing table entry The index is put into the distributed data structure and notifies NP together, thus helping NP to update the multi-level search tree. In this way, the traditional LPM algorithm implemented on one CPU is divided into two, that is, the longest matching is realized through the coordinated work of the two CPUs.

In order to maintain the routing prefix parent-child relationship structure, the control CPU needs to allocate additional valuable memory resources. In order to avoid this huge memory overhead, the present invention stores the network processor forwarding table index in the storage structure of the FIB table, so as to take into account the system software. Dual function of routing information table and NP parent-child relationship structure.

Description of drawings

Fig. 1 is the processing flowchart of route increase and route deletion of the present invention;

Fig. 2 is the storage structure of the FIB table on the control CPU of the present invention.

Detailed ways

Referring to Fig. 1 and Fig. 2, the technical solution of the present invention is described by taking the implementation of adding or deleting routes as an example.

1. The route adding process, the content of the routing table entry is as follows:

Destinatio IP: 192.168.10.1

Subnet Mask: 255.255.255.0

NexthopIp: 10.5.4.1

Out interface: Vlan10, Vlan ID: 10

There was no Np_forward_id field in the Fib_node structure, but now it is necessary to assist NP to implement LPM forwarding, so this field is added, calculated by NP and then notified to the control CPU to fill it.

Step 1: Control the CPU to perform a hash search based on the Next hop IP to obtain an AT index, which is an adjacency table index that links the next-hop host IP and Layer 2 egress information. This index will be notified to NP together with the delivered routing table entry.

Step 2: The control CPU searches the FIB table according to the prefix information, ie 192.168.10.0/24, to see if the routing item already exists, and if not, inform the NP to add a route.

Step 3: After receiving the message of adding a route, the NP first finds the corresponding entry in the AT table maintained by itself according to the AT index, then fills in the Layer 2 egress information, and then takes an available index from the forwarding table, and transfers the route to The content of the entry (route prefix, AT index, egress Vlan ID, number of next hops, etc.) is filled into it.

Step 4: NP determines all nodes in the search tree occupied by the route to be added in the three-level search tree structure according to the LPM algorithm. The LPM algorithm is an existing technology, and the specific algorithm is not described in detail in this paper. Next populate the forwarding table indexes into these nodes.

Step 5: NP notifies the control CPU of the forwarding table index.

Step 6: The control CPU fills the NP forwarding table index together with the entire routing table entry into the corresponding node of the software FIB table, and the route adding process ends.

In this way, when the routing item is regularly updated next time, the control CPU will first find it from the FIB table, and take out the stored NP forwarding table index therefrom, and then notify the NP to update. Next, the NP can directly update the content of the forwarding table and the content of the AT entry according to the index of the forwarding table.

2. In the route deletion process, it is assumed that the following two routes already exist in the FIB table:

(1) Prefix 192.168.10.0/24 Next hop Ip: 10.5.4.1 NP Forward table index: 10

(2) Prefix 192.168.10.0/22 Next hop Ip: 20.5.4.1 NP Forward table index: 20

Route (1) is now to be deleted.

And there are the following 4 routes in the 3-level search tree of NP:

<R1>Prefix 192.168.8.0/24 Next hop Ip: 20.5.4.1 NP Forward table index: 20

<R2>Prefix 192.168.9.0/24 Next hop Ip: 20.5.4.1 NP Forward table index: 20

<R3>Prefix 192.168.10.0/24 Next hop Ip: 10.5.4.1 NP Forward table index: 10

<R4>Prefix 192.168.11.0/24 Next hop Ip: 20.5.4.1 NP Forward table index: 20

According to the improved method mentioned above, the control CPU first finds the parent prefix information of route (1) in the FIB according to the subnet mask length, that is, Prefix 192.168.10.0/22, and obtains the forwarding table index 20 stored therein. Then notify the NP to delete the route. NP locates the routing prefix Prefix 192.168.10.0/24 in the 3-level search tree according to the LPM algorithm, because the control CPU notifies NP that there is a parent prefix, so NP cannot directly delete the route, but restores its parent prefix forwarding table index to search In the tree node, the index 20 is filled, so there are the following four routes in the three-level search tree:

<R1>Prefix 192.168.8.0/24 Next hop Ip: 20.5.4.1 NP Forward table id: 20

<R2>Prefix 192.168.9.0/24 Next hop Ip: 20.5.4.1 NP Forward table id: 20

<R3>Prefix 192.168.10.0/24 Next hop Ip: 20.5.4.1 NP Forward table id: 20

<R4>Prefix 192.168.11.0/24 Next hop Ip: 20.5.4.1 NP Forward table id: 20

From the above example of deleting a route, it can be seen that the forwarding module NP does not need to search for the parent prefix information, but can directly modify the level 3 search tree according to the message sent by the control CPU, which greatly simplifies the update process of the NP's routing table entry.

Claims (3)

1. A method of splitting the LPM algorithm to two CPUs, comprising: (1) two CPUs comprise network processor and control CPU, guide the control CPU that network processor routing table is set to complete the work of maintaining the parent-child relationship of routing prefix; (2) Before updating the routing table entry, first look up the FIB table on the control CPU to determine whether there is a prefix shorter than the current routing prefix, that is, the father prefix. If it exists, the mask length of the father prefix and the corresponding route The index of the entry is put into the distributed data structure, and the network processor is notified together to implement the LPM algorithm. 2. the method that LPM algorithm is split to two CPUs as claimed in claim 1 is characterized in that depositing network processor forwarding table index in the FIB table of control CPU. 3. the method that LPM algorithm is split into two CPUs as claimed in claim 2, is characterized in that described step (2) specifically comprises: (1) When adding a route, the control CPU first searches the FIB table. If the routing prefix already exists, it takes out the network processor forwarding table index stored in the FIB table, and notifies the network processor together with the routing table entry, and the network processor enters the update Process, if the routing prefix does not exist in the FIB table of the controlling CPU, the network processor is notified to enter the adding process, and the network processor uses the LPM algorithm to create a new forwarding entry, and write the entry index into the multi-level search tree In the node, the routing table item is quickly located in case of routing lookup, and the index is returned to the control CPU at the same time, and the control CPU writes the network processor forwarding table index together with the entire routing table item into its own FIB table

(2) When deleting a route, the control CPU first searches the FIB table to see if there is a shorter routing prefix, and if it exists, it takes out the father prefix information stored therein and informs the network processor, and the network processor indexes the father prefix forwarding table according to the LPM algorithm Revert to the search tree node.

Images