Article Preview
TopIntroduction
Cloud applications and services such as social networks, file sharing services, and file storage have become increasingly popular among users in recent years. For example, YouTube users download about 300 hours of video every minute and watch hundreds of millions of hours every day (Donchev, 2017). The growth of data volumes according to Cisco forecasts will double by 2021 compared with 2017, which requires the construction of new data centers (Cisco, 2018). Moreover, there is a tendency to concentrate processing and storing data in largescale data centers with thousands of servers and tens of thousands of virtual machines (VM).
Virtual machine migration is a transfer of a running virtual machine from one physical server to another. Migration is used in data centers to consolidate virtual machines on fewer physical servers, load balancing and reduce server overheating (Mishra, Das, Kulkarni, & Sahoo, 2012).
Despite the fact that the migration can be carried out without stopping the service, the deterioration of the performance of virtual machines and short-term downtime is still inevitable. In Clark et al. (2005), it was shown that the downtime can vary significantly from 60 ms to 3 seconds depending on the size of the virtual machine's RAM and how the application uses memory and network bandwidth. Moreover, the migration additionally loads the servers and network, which leads to lower application performance and affects the server monitoring process (Vorozhtsov, Toutova, & Toutov, 2018).
In the data center resource management system, migration can occur automatically in the following cases:
- •
Overheating of physical servers or increasing their temperature to a critical threshold;
- •
Lack of physical resources for VM operation;
- •
Consolidation of underloaded virtual machines on fewer physical servers.
It is known that in case of a non-optimally configured system, the total number of migrations increases and reverse migrations (ping-pong) appear. These can lead to instability of the data center resource management system. That is why more accurate models are needed to help datacenter administrators intelligently provision and control their virtualized infrastructure.
In order to VM migration does not affect the server monitoring process, the size of the monitoring window should be longer than migration time. At the same time, the monitoring window should not be too large, because of the risk of missing critical load changes, which can lead to a service level agreement (SLA) violation, as well as server overloading and overheating. Therefore, it is necessary to estimate the total migration time in order to set the optimal size of the sliding window for server monitoring.
The duration of migration is taken into account when developing requirements for computing and telecommunication systems of a cloud data center, as well as in calculating fault tolerance (Aleksankov, 2017). In addition, total migration time can be used in the formation of SLA-agreements for a certain class of applications.
The aim of this work is to develop a method of calculation the total migration time of virtual machines with various applications that are necessary to maintain the stability of migration and guarantee the quality of service indicators specified in the SLA-agreements. This paper summarizes the results obtained over a long period into a single article.