Abstract Virtual machine migration is used in cloud computing for server consolidation, system maintenance, opportunistic power savings, and load balancing. Scatter-Gather VM live migration is a new type of virtual machine migration. It is proposed to decouple the source quickly and reduce the time to evict the state of a migrating virtual machine from the host. In Scatter-Gather live migration, the memory pages of the VM are transferred not just to the destination but also to several pre-selected intermediaries. This open problem of selection of the intermediate nodes is not addressed in the literature. In this paper, we propose to define the problem of intermediate node selection in Scatter-Gather migration and prove that it is NP-complete. The problem is mathematically modelled as an integer programming problem based on two optimality criteria – minimizing eviction time and minimizing energy. Two heuristic algorithms – maximum-decrease-in-eviction-time and least-increase-in-energy are proposed to solve the problem and their performance is analyzed with respect to three parameters – eviction time, energy and total migration.

中文翻译：

---

云数据中心Scatter-Gather VM迁移的中间节点选择

摘要 虚拟机迁移在云计算中用于服务器整合、系统维护、机会节能和负载平衡。Scatter-Gather VM 实时迁移是一种新型的虚拟机迁移。建议快速解耦源并减少从主机驱逐虚拟机状态的时间。在 Scatter-Gather 实时迁移中，VM 的内存页面不仅会传输到目的地，还会传输到多个预先选择的中介。选择中间节点的开放性问题在文献中没有解决。在本文中，我们提出定义 Scatter-Gather 迁移中的中间节点选择问题并证明它是 NP 完全的。该问题在数学上被建模为基于两个最优性标准的整数规划问题——最小化驱逐时间和最小化能量。提出了两种启发式算法——最大减少驱逐时间和最小增加能量来解决这个问题，并针对驱逐时间、能量和总迁移三个参数分析它们的性能。