Devising a method for data consistency at replication in multicloud systems

This study’s object is the consistency of replication data in geo-distributed multicloud systems. The task under consideration is to devise a method for interval ordering of incoming requests by forming a sequence of general order numbers, according to which the order of writing data to geo-distributed replicas is executed. A feature of the proposed method is the a priori determination of equally long non-intersecting intervals of adjustment of incoming packet numbers, during which the incoming packet numbers are ordered. Grouping users into conditional clusters according to geographical location around brokers makes it possible to determine the priorities of sorting incoming packets. Brokers should be located near the leading replica of each cloud service provider and accept write requests from users from the nearest conditional cluster. In addition, the ordering of incoming packets occurs in the order of their arrival at each broker during the specified intervals. These mechanisms make it possible to synchronize data write operations in one step of global communication between geo-distributed replicas of separate cloud service providers. To estimate the latency of forming the total ordered sequence of incoming packet numbers, a simulation model has been built, whose feature is the ability to reproduce a different number and geographical location of replicas. The stability of a latency in coordinating incoming write requests in geo-distributed multicloud system with an increase in the intensity of the incoming flow by even 70 times has been shown experimentally. The results make it possible not only to reduce the latency of writing data to replicas of existing multicloud systems but also to choose the best geographical location of cloud service provider resources when designing new ones