What Is Adaptive Routing?

The internetwork router is a key component of a massively parallel processor system. According to the routing algorithm it uses, it can be divided into two types: deterministic and adaptive routers. Adaptive routers, for a pair of source and destination nodes, depend on the work of the network. Status, there are multiple paths to choose from.

Internet routers are massively parallel processors (MassivelyParallelProeessors,

PBFAA algorithm is a plane-based fully adaptive shortest wormhole routing algorithm (Planar-BasedFullyAdaptiveAlgorithm, PBFAA).

A lot of research has been done on adaptive routing algorithms using wormhole routing switching technology in direct networks. Many algorithms have been proposed, but they have the disadvantages of limited adaptiveness, high cost, or insufficient flexibility. Based on the existing algorithms, with low communication latency, high network throughput, and easy VLSI implementation as design goals, a fully scalable and adaptive plane-based fully adaptive routing algorithm PBFAA was proposed.

The algorithm divides the network into two virtual networks VIN0 and VI1I. The virtual channel in VlN0 routes messages according to the plane adaptive routing policy. The virtual channel in VIN1 can completely adaptively route messages. VIN0 guarantees the deadlock-free nature of the algorithm. Both networks are adaptive, so compared with existing better algorithms such as (channel), the algorithm is more adaptive, and it can fully and effectively utilize network resources, improve network throughput, and fault tolerance. More powerfully, the following introduces the PBFAA algorithm with an n-dimensional mesh network:

1) The algorithm sets 4 virtual channels for each physical channel, which are represented by VC _{dimension, label, and direction} , where dimension indicates the dimension along which the virtual channel passes messages; label indicates the serial number of the virtual channel, and the value is 0, 1, 2 or 3; direction can be + (meaning the message will be passed in the forward direction) or-(meaning the message will be passed in the reverse direction). For example, VC¨, a negative virtual channel with a sequence number 1 on the first dimension of a node.

2) Divide the network into two virtual networks: VIN0 and VIN1. A virtual channel with a serial number from 0 to 2 is used in VIN0; a virtual channel with a serial number 3 is used in VIN1.

3) In VIN0, select the virtual channel routing message to the destination node according to the plane adaptive routing strategy; in VIN1, select the virtual channel routing message to the destination node according to the fully adaptive shortest path routing strategy. The virtual channels that can be selected according to the corresponding routing policies in the two virtual networks are called required virtual channels, and the idle required virtual channels are called available virtual channels.

4) When the header chip of a message reaches a certain node, if the node is the destination node, the message is received, otherwise:

a) If there are available virtual channels, output the virtual channel selection strategy based on the maximum spacing of available virtual channels, and apply for Req to the corresponding dimension virtual channels; if there are no available virtual channels, suspend the application and wait until the required virtual channel changes Make the same application as available;

b) If the application is responded, the message is transmitted to the neighboring node along the corresponding virtual channel; if the application is not responded, the operation of the above a) is re-performed at the next beat until the application is responded to the message. Node. Repeat the above operations on each intermediate node until the message is passed to the destination node. The so-called maximum distance output virtual channel selection strategy means that among the channels allowed to be accessed, an application is first applied to the virtual channel with the largest absolute value of the dimension distance (middle node to destination node) of the dimension in which the path is to be reduced.

The planar adaptive routing strategy used in VIN0 is: in an n-dimensional mesh network, sort the virtual channels of each physical channel, and use C _{i, j to} represent the set of all virtual channels with the sequence number j in the i-th dimension. , It can be divided into forward virtual channel set C _{i, j +} and reverse virtual channel set C _{i, j-} plane adaptive

The routing algorithm defines n-1 adaptive planes A ₀ to A _{n-1, and} each plane is composed of adjacent two-dimensional virtual channels: A _i = C _{i, 0} + C _{i + 1,1} + C _{i + 1,2 ,} 0in-2. The algorithm can be divided into two levels (high-level and low-level):

1. High-level algorithm: (between adaptive planes)

   1) For i = 0, i <(m-1), i ++ do

   Adaptively route messages to the destination node in the A plane (see low-level algorithms)

   end.

   2) After the above process ends, if the message has not reached the destination node, the message is routed to the destination node through the virtual channel in A _n-2 = C _n-1,0 .

2. Low-level algorithm (in the adaptive plane):

The adaptive plane A contains the virtual channel set C _{i, 0} , C _{i + 1,1} and C _{i + 1,2.} In A, the message approaches the destination node in the i-th and i + 1-dimension and routes adaptively. . To avoid deadlocks, messages are divided into two types. One type is called a direction-increasing message that needs to add the i-dimensional address during the routing process, and the other type is a direction-reducing message that needs to reduce the i-dimensional address. At the same time, the virtual channel in A is divided into two separate virtual subnets: the increasing subnet (including the virtual channel set C _{i, 0 +} and C _{i + 1,1} ) and the decreasing subnet (including the virtual channel set C) _{i, 0-} and C _{i + 1,2} ). In this way, the increasing message is routed on the increasing subnet, and the decreasing message is routed on the decreasing subnet. Each message can adaptively approach the destination node in the corresponding subnet. When the i-dimensional address of the intermediate node reached by the message is equal to the i-dimensional address of the destination node, the routing process in A ends, and the process moves to the next high-level step. ^[1]

The routing strategy in multi-layer satellite networks is different from traditional static routing. Although the constellation topology is periodic and deterministic, dynamic routing can dynamically maintain routing tables and exchange routing information in a timely manner. The adaptive routing strategy needs to know the length and traffic on each ISL in the network, and adaptively select the optimal path that meets the requirements for effectiveness and reliability. The route calculation applies the discretization method to calculate the route and update the routing table at each fixed time t _k , t _k = kt, k = 0,1,2, ......, K1 and consider that in the time interval t The internal network topology is unchanged, and the routes are constantly switched using a fixed-time switching strategy when the routing table is updated.

The following assumptions are required for network analysis:

1. Network satellites are independent of each other;

2. Each satellite node bears different service loads according to the service model;

3. The amount of independent traffic carried by each satellite node is related to the size and geographic location of the satellite coverage; ^[2]

Adaptive routing routing algorithm

The routing algorithm uses the Bellman-Ford backward routing algorithm to determine the optimal path. The path's comprehensive weight (TPW, TPW) represents the comprehensive performance of a path's delay and bandwidth occupation. It also considers effective and reliable comprehensive performance. TPW consists of three parts, uplink delay D _u , downlink delay D _d and the link weight LW _wi of each ISL _wi on the path _{, which} represents the set of ISLs on the path W = {w ₁ , w ₂ , ......., w _i , ......, w _ns-1 }, | W | = n _s 1 means that the path contains n _s 1 ISL, and n _{s is on} the path Of satellites (including source and target satellites). Among them, after the ground source and target positions are determined, the source satellite and target satellite are selected using the maximum elevation access scheme.

Among them, D _wi represents the transmission delay of ISL _wi , W _wi represents the average on-board processing and exchange delay, and f is the information weight parameter. The solution of D _u , D _d and D _wi only need to know the satellite space position coordinates, and divide the link length by the transmission speed without redundancy.

According to Jackson's principle, for data packet services, each ISL can be regarded as a single service window hybrid queuing model M / M / 1 / m. The interval between the arrival of data packets follows a negative exponential distribution with the parameter ; the service time is the parameter As is a negative exponential distribution, each ISL has m data packet queuing capacity. When there are m data packets in the system, the new data packets no longer enter the queue. Packet is dropped, there is

= /

The average packet processing and switching delay is ^[2]

Steps in adaptive routing strategy

The routing strategy steps in a multi-layer satellite network are as follows:

Step 1 Set basic parameters and network initialization

Step 2 fix a time t _k, in this time interval t solve the satellite orbit parameters, calculate the satellite position coordinates and ISL length, and establish the network topology

Step 3 Calculate the ISL load of the MLSN according to the set business model

Step 4 According to the queuing theory, calculate the delay of processing / exchanging on the satellite of the data packet, as shown in equation (14)

Step 5 Find source satellites and target satellites (LEO, MEO, and GEO source / target satellites) in each satellite layer based on the ground source / target location

Step 6 According to QOS requirements and network status, select the satellite layer for transmitting services, and find the optimal path according to the Bellman-Ford routing algorithm. The default service bearer satellite layer is the LEO layer, but if the MEO source / target satellites are the same and the LEO source If the target satellites are different, go to Step 9; if the GEO source / target satellites are the same, and LEO and MEO source / target satellites are different, go to Step 10; otherwise, go to Step 7

Step 7 If the LEO layer path contains the number of ISLs ISL _LEO threshold, go to Step 8; if the LEO layer path contains the number of ISLs> ISL _LEO threshold, and the MEO layer path contains ISL numbers ISL _LEO threshold, go to Step 9; otherwise, go to Step 10

Step 8 Establish the optimal communication path of the LEO satellite layer to complete the transmission task

Step 9 Establish the optimal communication path of the MEO satellite layer to complete the transmission task

Step 10 Establish the optimal communication path of the GEO satellite layer to complete the transmission task

Step 11: Calculate the characteristic parameters of the multilayer network and analyze the network performance

Step 12 Update the time interval, complete the calculation of the new routing table, and complete the handoff of the satellite ^[2]

Adaptive routing feature

The multi-layer satellite network adaptive routing strategy has the following characteristics: the service accesses the satellite system through the LEO source satellite and the target satellite, and selects the satellite layer that transmits the service according to the QOS needs and network status. If the LEO layer network resources cannot meet the service requirements This service is transferred to the MEO layer transmission or even the GEO layer transmission. For terrestrial sources / targets, directly access the MEO or GEO satellite. Because the routing algorithm is simple to implement, it has not been analyzed in detail. In addition, the simulation results show that if the path of the LEO layer contains 6 or 7 ISLs, the delay will be greater than 200ms. At this time, if the service is transferred to MEO, the transmission time is shorter and less resources are consumed on the satellite. And if the ground source and target locations are covered by the same MEO or GEO satellite, then the service is transferred to MEO and GEO transmission to reduce the occupation of resources on the satellite. This strategy considers the delay index and ISL bandwidth occupancy, and the optimal path selection takes into account the effectiveness and reliability of the satellite system. ^[2]