Load balancing method of server cluster

Abstract

The invention provides a load balancing method of a server cluster, and the method comprises the following steps: S1, judging whether the server cluster relates to a microservice architecture; S2, judging whether capacity adjustment is needed for server overall load; S3, receiving a request, and judging a request type according to feature information in the request; S4, inquiring and locating partitions of corresponding request types on a hash ring, and inquiring current load of related nodes, obtaining the current load of the node with the lowest load and obtaining a load mean value of the nodes on the partition; S5, performing calculation for the requests through a hash function according to classification characteristics, and mapping a result to a corresponding position of the partition; S6, according to a screening direction, judging whether a neighboring node is the node with the lowest load; S7, judging whether a current position is at a peak of a block. The method provided by the invention effectively balance load of each node in a system in an allocation stage and improves system efficiency.

Description

technical field [0001] The invention relates to a consistent hash allocation method, in particular to a load balancing method executed between service nodes of a server cluster, and belongs to the field of computer algorithms. Background technique [0002] Load balancing is widely used in distributed environments, including big data processing in the field of data processing. For example, in Map / Reduce of Hadoop platform, load balancing is used to alleviate the data skew problem on the Reduce side; user load balancing of traditional P2P servers; distributed service systems request processing load balancing. [0003] Nowadays, due to the wide application of big data and cloud computing, the application of distributed service systems is becoming more and more frequent. Due to the need for horizontal expansion of system performance, consistent hashing has become a more mature solution in this regard. The essence of consistent hashing is to use a hash ring instead of the tradit...

AI Technical Summary

Problems solved by technology

[0006] 1. For requests with the same characteristic string, it is easy to get the same result, which makes the randomness of the result not obvious, causing some nodes to carry too many requests, while the load of other nodes is too low, and the processing efficiency of the whole system deviates from the ideal seriously state

[0007] 2. The lack of a suitable load balancing strategy cannot effectively deal with the excessive load on some nodes, and naturally the optimal adjustment of the system cannot be implemented

[0008] 3. It cannot adapt to the new type of distributed system, especially when the services provided by each node are not the same, it needs to pay an additional price - multi-layer hash ring and other methods to solve the situation that the node does not correspond to the service content

The traditional way is to partially alleviate the problem of skew by splitting a node into the original node and another virtual node (virtual node) that is opposite to it on the ring. To alleviate the tilt problem, not only is the efficiency low, but its effect is very limited

Images