Load balancing algorithm in a distributed programming system, built on the basis of the HDP protocol

In this paper, we discuss load balancing in a software system based on the HDP protocol. Load balancing is a method of distributing requests between several identical nodes of a distributed computing system in order to optimize the use of all nodes together. There are various load balancing algorithms that have different applications in different areas and their purpose is also to ensure the fault tolerance of the system as a whole. The algorithm of round robin is a cyclic search of all nodes in a circle, each subsequent request is sent to the next node in the set. The main advantage is the simplicity of the algorithm, which makes it possible to use it in many systems and the architecture of such systems remains as simple as possible. The operational characteristics of the system do not add to the complexity of the architecture. From the disadvantages of the algorithm is to distinguish the uneven distribution of the load, if the nodes of the network have different technical characteristics. A technically more powerful node will better cope with requests than others, which will be more loaded. Also, the problem of an inaccessible node is not solved, i.e. which during the operation of the system suddenly went out of order. The algorithm weighted round robin is a modification of the round robin algorithm, in which each node is manually assigned a certain weight - the probability with which the load balancer next time will choose this particular node. Compared with the round robin in this algorithm, it was possible to compensate for possible technical differences between the nodes of the network. From the disadvantages of the algorithm, it is worth highlighting the lack of a solution to all fault-tolerance problems, because if a network node with a high priority ceases to be available, queries are still sent to it. The main problem with most balancing algorithms is that they do not use the state information of each node that is in the list from which the balancer selects the appropriate node each time based on a certain algorithm. The feedback node algorithm proposed by the author is an algorithm for load distribution in a distributed system based on the HDP protocol. This algorithm takes into account heterogeneous data about the current load of each node and, based on these data, distributes the load between them. An agent is installed on the monitored node, which transmits to the server the data on the overall load, the processor load, and the use of memory on this node. The HDP balancer uses this data to decide whether to send a request to a particular node, depending on its load at the current time.

Keywords

Authors

References