Gusev E. Techniques of organizations shared access to distributed memory pages in cloud computing systems

The thesis is devoted to the problem of common resource processing optimization in a cloud computing systems. There is proposed access technique to shared resource, which allows exclude the overloads, associated with locks in the cloud environment and reduce the requirements for data transfer channels. The paper substantiates the need to reduce response time for reducing overloading probability. To solve the problem are used three methods: the method of two-phase transaction execution, method of piping phases and method of resource handler assigning. The first two use cached old versions of pages to determine required list of resources during first phase, and reexecutes transaction on second phase to confirm ACID. The piping phase method is more efficient for transactions with low cached resources due to piping approach. Method of resource handler assigning reduces queues to intensively updated resource pages. The method idea - to fix for each "hot" page processing node, thereby minimizing the number of network hops during page access. To estimate the effect of the method the author offers three-component mathematical model consisting of: delay models to determine the duration of access to memory page, the model of conflict - to evaluate the effect of blocking, and traffic model - to synthesize traffic.