KR102153814B1 - Stochastic Routing Algorithm for Load-balancing Interconnection Network System - Google Patents

Abstract

A probability-based adaptive routing method and system for load balancing is presented. In the probability-based adaptive routing method for load balancing proposed in the present invention, a destination is identified by receiving a packet, and a minimum path and a minimum proximity path for a transmission path from the source to the destination are calculated, and the calculated queue for each path And calculating a length, calculating a selection probability according to a load for each path, and transmitting a packet by selecting a path according to the calculated selection probability.

Description

Probability-based adaptive routing algorithm for load balancing {Stochastic Routing Algorithm for Load-balancing Interconnection Network System}

The present invention relates to a probability-based adaptive routing method and system for load balancing.

The technology for adaptive route designation using aggregate communication congestion information (US 2018-0006950) is a congestion value that can be obtained based on local occupancy and remote occupancy values inside the switch. As a technique, a packet is transmitted to an output port with the minimum congestion based on the congestion value obtained above.

The mechanism for controlling resource utilization using adaptive routing (Korean Patent Laid-Open Patent No. 10-2016-0004348) is a technology that uses an adaptive routing algorithm based on Content Aware Adaptive Throttle (CAAT), and is a deterministic routing. Among the determined path and the adaptive routing path, a path that does not cause congestion is selected.

A technique for adaptive path designation using aggregate communication congestion information transmits all packets in one path with minimal congestion, and thus a bottleneck and load imbalance problem arise.

The mechanism for controlling resource utilization using adaptive routing does not use adaptive routing but uses deterministic routing when the load of the adaptive routing path is over a certain level to alleviate the above problem. Even in the case of a mechanism for controlling resource utilization using adaptive routing, since packets are transmitted only through the adaptive routing path and the deterministic path, there is a problem in that equipment existing inside the network system cannot be sufficiently utilized.

The technical problem to be achieved by the present invention is to provide a method and system for efficiently distributing the load throughout the network through probabilistic route selection based on the current load situation of each route by considering two or more routes between a source and a destination. There is.

In one aspect, the probability-based adaptive routing method for load balancing proposed in the present invention comprises the steps of receiving a packet, identifying a destination, and calculating a minimum path and a minimum proximity path for a transmission path from a source to a destination, And calculating a calculated queue length for each path, calculating a selection probability according to a load for each path, and selecting a path according to the calculated selection probability and transmitting a packet.

In the step of receiving a packet to confirm the destination, and calculating the minimum path and the minimum proximity path for the transmission path from the source to the destination, after determining the minimum path set and the minimum proximity path set from one node to another node, The loads for all nodes on each path in the path set and the minimum proximity path set are updated every predetermined time, and the sum of all nodes in the path for each path is defined as the load of the corresponding path.

Calculating the calculated queue length for each path and calculating the selection probability according to the load for each path is to ensure that the load of each path is balanced when the packet is transmitted from one node to another, so that the load of each path is small. Calculate the probability of selection with higher probability.

In the step of selecting a path according to the calculated selection probability and transmitting a packet, a probabilistic routing method is used to distribute the load by selecting a path with a higher probability as the load of each path is less in order to reduce the bottleneck.

In another aspect, the probability-based adaptive routing system for load balancing proposed in the present invention receives a packet, identifies a destination, and calculates a minimum path and a minimum proximity path for a transmission path from a source to a destination. A path search unit, a probability calculation unit that calculates a calculated queue length for each path, a selection probability according to a load for each path, and a packet transmission unit that selects a path according to the calculated selection probability and transmits a packet.

The path search unit determines a set of minimum paths from one node to another and a set of minimum proximity paths, and then updates the loads of all nodes on each path belonging to the set of minimum paths and minimum proximity paths at predetermined times. And, for each path, the sum of all nodes in the path is defined as the load of the path.

When transmitting a packet from one node to another node in order to balance the load of each path, the probability calculator calculates a selection probability having a higher probability as the load of each path is less.

In order to reduce the bottleneck, the packet transmission unit uses a probabilistic routing method in which the load is distributed by selecting a path with a higher probability as the load of each path decreases.

According to embodiments of the present invention, the load can be efficiently distributed throughout the network through probabilistic path selection based on the current load situation of each path in consideration of two or more paths between a source and a destination. In addition, it is possible to efficiently use the equipment inside the network system, and accordingly, the packet processing speed performance can be improved.

1 is a diagram illustrating a probability-based adaptive routing algorithm for load balancing according to an embodiment of the present invention.
2 is a flowchart illustrating a probability-based adaptive routing method for load balancing according to an embodiment of the present invention.
3 is a diagram illustrating a configuration of a probability-based adaptive routing system for load balancing according to an embodiment of the present invention.

The present invention proposes a high performance computing system (HPC) and a method of probabilistically determining routing paths from one node to another in a data center by placing multiple paths as candidates. A data center refers to a central processing unit that manages large-scale servers and databases by connecting hundreds to thousands of computers through a switch.

In the present invention, the load of a node is defined as follows. The dictionary meaning is'the number of packets to be processed that have not yet been processed by the node', and can also be defined as'expected waiting time until leaving the node when a packet arrives at the node'. After determining the minimum set of paths from one node to the other and the set of paths that are close to the minimum, the loads for all nodes on each path belonging to the two sets are updated every certain time, and for each path, The sum of all nodes' is defined as the load of the path. Now, in transmitting packets from one node to another node, the smaller the load value of each path, the higher the probability of selecting a path and setting it as a routing path. Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating a probability-based adaptive routing algorithm for load balancing according to an embodiment of the present invention.

The adaptive routing algorithms in the prior art select another minimum path when the optimal path is in a congestion state. At this time, a problem of load imbalance may occur.

를 선택할 확률

는 각 경로의 로드를 고려하여 결정한다. 각 경로의 로드가 균형을 이루도록 하는

설정을 통한 확률적 라우팅으로, 종래기술의 적응형 라우팅 알고리즘들의 로드 불균형 문제를 해결할 수 있다. According to an embodiment of the present invention, the k-th path among k paths for an arbitrary source-destination pair

The probability of choosing

Is determined by considering the load of each path. To balance the load on each path

With probabilistic routing through configuration, it is possible to solve the load imbalance problem of adaptive routing algorithms of the prior art.

2 is a flowchart illustrating a probability-based adaptive routing method for load balancing according to an embodiment of the present invention.

Probability-based adaptive routing method for load balancing is the step of receiving a packet to confirm a destination, calculating the minimum path and the minimum proximity path for the transmission path from the source to the destination (210), the calculated queue for each path It includes a step 220 of calculating a length, calculating a selection probability according to a load for each path, and a step 230 of selecting a path according to the calculated selection probability and transmitting a packet.

In step 210, the packet is received (211) to check the destination (212), and the minimum path and the minimum proximity path for the transmission path from the source to the destination are calculated (213). After determining the minimum path set and the minimum proximity path set from one node to another node, the load for all nodes on each path belonging to the minimum path set and the minimum proximity path set is updated every predetermined time, and each For a path, the sum of all nodes in the path is defined as the load of the path.

를 계산한다(222). 각 경로의 로드가 균형을 이루도록 하기 위해 하나의 노드에서 다른 노드로 패킷을 전송할 때, 각 경로의 로드가 적을 수록 높은 확률을 갖는 선택 확률을 계산한다. In step 220, the calculated queue length for each path is calculated (221), and the selection probability according to the load for each path

Calculate (222). When a packet is transmitted from one node to another in order to balance the load of each path, the smaller the load of each path, the higher the probability of selection is calculated.

설정 과정에 대하여 설명한다. For load balancing according to an embodiment of the present invention

The setting process will be described.

In total k paths between source and destination, when the minimum path is N and the minimum proximity path is M, it can be expressed as follows:

N minimum paths

M smallest proximity paths

를 선택할 확률

는 다음과 같이 나타낼 수 있다: At this time, the k-th path

The probability of choosing

Can be represented as:

는 경로

의 큐 길이이고, T는 큐 길이 업데이트의 주기이다. here,

The path

Is the queue length, and T is the period of the queue length update.

는 다음과 같이 나타낼 수 있다: Load of the kth path at this time

Can be represented as:

에 따른 경로를 선택(231)하여 패킷을 전송한다(232). 병목현상을 감소시키기 위해 각 경로의 로드가 적을 수록 높은 확률로 경로를 선택하여 로드를 분산시키는 확률론적 라우팅 방법을 이용한다. In step 230, the calculated selection probability

The packet is transmitted by selecting a path according to (231) (232). In order to reduce the bottleneck, a stochastic routing method is used to distribute the load by selecting a path with a higher probability as the load of each path decreases.

That is, the present invention is a probability-based routing method that can dramatically improve the performance of a deterministic routing method.

In addition, while several existing routing methods may be limited to several specific topologies and used, the present invention is a routing technique applicable to all topologies.

Therefore, the routing scheme according to an embodiment of the present invention can greatly reduce a bottleneck that may occur when data is transmitted between nodes inside a supercomputer/data center, and thus can have a fault-tolerant feature. have.

3 is a diagram illustrating a configuration of a probability-based adaptive routing system for load balancing according to an embodiment of the present invention.

The probability-based adaptive routing system 300 for load balancing according to the present embodiment may include a processor 310, a bus 320, a network interface 330, a memory 340, and a database 350. The memory 340 may include an operating system 341 and a probability-based adaptive routing routine 342 for load balancing. The processor 310 may include a path search unit 311, a probability calculation unit 312, and a packet transmission unit 313. In other embodiments, the probability-based adaptive routing system 300 for load balancing may include more components than those of FIG. 3. However, there is no need to clearly show most of the prior art components. For example, the probability-based adaptive routing system 300 for load balancing may include other components such as a display or a transceiver.

The memory 340 is a computer-readable recording medium and may include a permanent mass storage device such as a random access memory (RAM), read only memory (ROM), and a disk drive. Also, the memory 340 may store program code for the operating system 341 and the probability-based adaptive routing routine 342 for load balancing. These software components may be loaded from a computer-readable recording medium separate from the memory 340 using a drive mechanism (not shown). Such a separate computer-readable recording medium may include a computer-readable recording medium (not shown) such as a floppy drive, a disk, a tape, a DVD/CD-ROM drive, or a memory card. In another embodiment, software components may be loaded into the memory 340 through a network interface 330 rather than a computer-readable recording medium.

The bus 320 may enable communication and data transmission between components of the probability-based adaptive routing system 300 for load balancing. Bus 320 may be configured using a high-speed serial bus, a parallel bus, a storage area network (SAN), and/or other suitable communication technology.

The network interface 330 may be a computer hardware component for connecting the probability-based adaptive routing system 300 for load balancing to a computer network. The network interface 330 may connect the probability-based adaptive routing system 300 for load balancing to a computer network through a wireless or wired connection.

The database 350 may serve to store and maintain all information necessary for probability-based adaptive routing for load balancing. In FIG. 3, the database 350 is constructed and included in the probability-based adaptive routing system 300 for load balancing, but is not limited thereto and may be omitted depending on the system implementation method or environment, or It is also possible for all or some of the databases to exist as external databases built on separate and different systems.

The processor 310 may be configured to process instructions of a computer program by performing input/output operations of the probability-based adaptive routing system 300 for basic arithmetic, logic, and load balancing. Instructions may be provided to the processor 310 by way of memory 340 or network interface 330 and via bus 320. The processor 310 may be configured to execute program codes for the path search unit 311, the probability calculation unit 312, and the packet transmission unit 313. Such program codes may be stored in a recording device such as the memory 340.

The path search unit 311, the probability calculation unit 312, and the packet transmission unit 313 may be configured to perform the steps 210 to 230 of FIG. 2.

The probability-based adaptive routing system 300 for load balancing may include a path search unit 311, a probability calculation unit 312, and a packet transmission unit 313.

The path search unit 311 receives the packet, identifies a destination, and calculates a minimum path and a minimum proximity path for a transmission path from the source to the destination. After determining the minimum path set and the minimum proximity path set from one node to another node, the load for all nodes on each path belonging to the minimum path set and the minimum proximity path set is updated every predetermined time, and each For a path, the sum of all nodes in the path is defined as the load of the path.

를 계산한다. 각 경로의 로드가 균형을 이루도록 하기 위해 하나의 노드에서 다른 노드로 패킷을 전송할 때, 각 경로의 로드가 적을 수록 높은 확률을 갖는 선택 확률을 계산한다. The probability calculation unit 312 calculates the calculated queue length for each path, and the selection probability according to the load for each path

Calculate When a packet is transmitted from one node to another in order to balance the load of each path, the smaller the load of each path, the higher the probability of selection is calculated.

에 따른 경로를 선택하여 패킷을 전송한다. 병목현상을 감소시키기 위해 각 경로의 로드가 적을 수록 높은 확률로 경로를 선택하여 로드를 분산시키는 확률론적 라우팅 방법을 이용한다. The packet transmission unit 313 is the calculated selection probability

The packet is transmitted by selecting a path according to the following. In order to reduce the bottleneck, a stochastic routing method is used to distribute the load by selecting a path with a higher probability as the load of each path decreases.

The routing scheme according to an embodiment of the present invention can greatly reduce a bottleneck that may occur during data transmission between nodes in a supercomputer/data center, and thus can have a fault-tolerant feature.

The present invention can be adopted and used as a routing technique of a supercomputer/data center in all types of topology types that have been implemented in the past, and thus has an advantage of high versatility compared to several existing routing patents. In addition, when replacing the existing routing scheme with the routing proposed in the present invention, a bottleneck can be greatly reduced in the case of a situation where packets are concentrated in a specific node. As the number of nodes with bottlenecks increases, traffic does not arrive on time, packets are lost, and the number of retransmissions increases, which adversely affects the overall system performance.Thus, reduction of bottlenecks will greatly improve overall system performance. I can. Therefore, it is expected that when many companies implementing existing supercomputers/data centers adopt the technology proposed in the present invention, they will be able to significantly improve performance.

The apparatus described above may be implemented as a hardware component, a software component, and/or a combination of a hardware component and a software component. For example, the devices and components described in the embodiments include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It may be implemented using one or more general purpose computers or special purpose computers, such as a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications executed on the operating system. In addition, the processing device may access, store, manipulate, process, and generate data in response to the execution of software. For the convenience of understanding, although it is sometimes described that one processing device is used, one of ordinary skill in the art, the processing device is a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of these, configuring the processing unit to behave as desired or processed independently or collectively. You can command the device. Software and/or data may be interpreted by a processing device or to provide instructions or data to a processing device, of any type of machine, component, physical device, virtual equipment, computer storage medium or device. Can be embodyed in The software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -A hardware device specially configured to store and execute program instructions such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler but also high-level language codes that can be executed by a computer using an interpreter or the like.

Although the embodiments have been described by the limited embodiments and drawings as described above, various modifications and variations are possible from the above description to those of ordinary skill in the art. For example, the described techniques are performed in an order different from the described method, and/or components such as a system, structure, device, circuit, etc. described are combined or combined in a form different from the described method, or other components Alternatively, even if substituted or substituted by an equivalent, an appropriate result can be achieved.

Therefore, other implementations, other embodiments, and claims and equivalents fall within the scope of the claims to be described later.

Claims (8)

Receiving a packet to confirm a destination, and calculating a minimum route and a minimum proximity route for a transmission route from a source to a destination through a route search unit;
Calculating a calculated queue length for each path, and calculating a selection probability according to a load for each path through a probability calculator; And
Selecting a path according to the calculated selection probability and transmitting a packet through a packet transmission unit
Including,
Calculating the calculated queue length for each path, and calculating the selection probability according to the load for each path through a probability calculator,
The load of each path is defined using the number of packets to be processed by the node, and the expected waiting time until leaving the node when the packet arrives at the node, and the minimum set of paths from one node to another node and After determining the set of minimum proximity paths, update the loads for all nodes on each path belonging to the two sets, and define the sum of the loads for all nodes in the path for each path as the load of the path.
In order to balance the load of each path, when transmitting a packet from one node to another, the lower the load of each path, the higher the probability of selection is calculated.
Adaptive routing method. The method of claim 1,
Receiving a packet to confirm a destination, and calculating a minimum route and a minimum proximity route for a transmission route from the source to the destination through the route search unit,
After determining the minimum path set and the minimum proximity path set from one node to another node, the loads for all nodes on each path belonging to the minimum path set and the minimum proximity path set are updated every predetermined time, and each For a path, the sum of all nodes in the path is defined as the load of the path.
Adaptive routing method. delete The method of claim 1,
Selecting a path according to the calculated selection probability and transmitting a packet through the packet transmission unit,
In order to reduce the bottleneck, the less the load of each path, the higher the probability that the path is selected and the load is distributed using a probabilistic routing method.
Adaptive routing method. A path search unit that receives the packet, identifies a destination, and calculates a minimum path and a minimum proximity path for a transmission path from the source to the destination;
A probability calculator that calculates the calculated queue length for each path and calculates a selection probability according to the load for each path; And
A packet transmission unit that transmits a packet by selecting a path according to the calculated selection probability
Including,
The probability calculation unit,
The load of each path is defined using the number of packets to be processed by the node, and the expected waiting time until leaving the node when the packet arrives at the node, and the minimum set of paths from one node to another node and After determining the set of minimum proximity paths, update the loads for all nodes on each path belonging to the two sets, and define the sum of the loads for all nodes in the path for each path as the load of the path.
In order to balance the load of each path, when transmitting a packet from one node to another, the lower the load of each path, the higher the probability of selection is calculated.
Adaptive routing system. The method of claim 5,
The route search unit,
After determining the minimum path set and the minimum proximity path set from one node to another node, the loads for all nodes on each path belonging to the minimum path set and the minimum proximity path set are updated every predetermined time, and each For a path, the sum of all nodes in the path is defined as the load of the path.
Adaptive routing system. delete The method of claim 5,
The packet transmission unit,
In order to reduce the bottleneck, the less the load of each path, the higher the probability that the path is selected and the load is distributed using a probabilistic routing method.
Adaptive routing system.

Images