A Layout Method of Copy Selection Nodes on Network Devices Based on ILP Model

Abstract

The invention discloses a layout method of replica selection nodes on network equipment based on an ILP model, and is oriented to a distributed key value storage system located in a data center, wherein the data center uses a multi-layer network topology based on a tree structure. The layout method comprises the steps of dividing requests into different flow groups according to the sources of keyvalue storage requests; collecting flow statistical information of each flow group at the outlet of a switch at the top of each rack; and according to the flow statistical information, solving an ILPmodel to obtain the layout of replica selection nodes in a network. By dividing the requests of a key value storage system into different flow groups, determining the optimization objective and limiting conditions of the layout problem, and formalizing the layout problem into a linear integer programming model problem, the method can effectively reduce the occurrence of "a herd effect" in replicaselection, so that each replica selection node can select a replica according to a more recent system state, thereby improving the replica selection efficiency and reducing the response delay of the distributed key value storage system.

Description

technical field [0001] The invention belongs to the field of distributed key-value storage, and more specifically relates to a layout method of copy selection nodes on network equipment based on an ILP model. Background technique [0002] Distributed key-value stores are a core component in modern web applications. Considering the interactive nature of modern web applications, it is important to reduce the response latency of key-value stores. Since even one end-user request can generate hundreds or thousands of storage accesses, the mere high tail latency of a key-value store can have a huge impact on user-perceived latency. [0003] In order to achieve high availability and high reliability, distributed key-value storage usually stores data copies on multiple servers. A read request can fetch data from any replica. However, due to frequent performance fluctuations of the server, replica selection will directly affect the response delay of read requests. Considering tha...

AI Technical Summary

Problems solved by technology

Among them, "herd behavior" refers to the situation in which multiple RSNodes (Replica Selection Nodes) select a copy at the same time, resulting in a decrease in the performance of the copy

[0005] However, the network topology of modern data centers is very complex, and the layout of replica selection nodes on network devices will have a great impact on the effect of replica selection

Images