| With the development of cloud computing,the scale and number of cloud data centers are also expanding.At the same time,the number of users using cloud computing services has also surged.As one of the important carriers of user services,the number of virtual machines(VM)is also increasing,and a large number of heterogeneous VM are dynamically created and deregistered,resulting in a certain resource fragmentation during the operation of the data center.At the same time,with the proposal of the cloud-native concept and the development of domestic ARM chips,more and more data centers use multi-NUMA servers as computing resources to provide services to users.However,the use of NUMA server will exacerbate the problem of resource fragmentation in the data center due to the characteristics of resource segmentation.At the same time,cross-die or cross-chip memory access will affect the performance and life cycle of VM,introducing greater challenges to VM scheduling.In the online placement scenario of VM,to ensure the efficiency of VM scheduling,some greedybased heuristic algorithms are commonly applied.And,these Algorithms are often used to obtain the preferred optimal solution while making the decision,without considering the adverse impact of resource fragmentation caused by the current decision for the future scheduling.In addition,online placement algorithms alone are often unable to further improve data center resource utilization,especially in complex industrial application scenarios.How to further optimize deployment and VM consolidation under multiple constraints,reduce data center fragmented resources composed of NUMA servers,and improve the resource utilization is a great challenge.Given the above problems and challenges,this thesis studies two scenarios of virtual machine online placement and virtual machine consolidation for NUMA systems.The main research contents and contributions of the thesis are as follows:(1)Aiming at the virtual machine online placement for the NUMA server system,a nonlinear integer programming model is established considering the placement constraints of multiple NUMA nodes,a novel algorithm integrating multiple heuristic factors is proposed to solve the problem.The purposed algorithm not only considers the improvement of resource utilization but also considers the balance of multi-resource dimensions and the characteristics of access delay between multiple NUMA nodes,which can effectively improve resource utilization and reduce scheduling failures under the condition of limited performance impact.(2)A mathematical modeling of virtual machine consolidation for the heterogeneous NUMA server system in complex scenarios is carried out,which combines more practical industrial needs,not only considering NUMA resources and non-NUMA resource constraints but also anti-affinity service guarantee constraints.Meanwhile,a hybrid heuristic gray wolf algorithm is proposed to consolidate VM,effectively employ the resource allocation of VM in multiple NUMA server clusters,and balance the two cost goals of the number of activated servers and the number of VM migrations,achieving a good trade-off.(3)Based on Cloud Sim Py,a NUMA server data center scheduling management simulation platform Numa Sim was developed.Combined with the Microsoft Azure Trace for Packing 2020 data set and the real NUMA data center data set provided by well-known cloud service providers,we construct some synthetic data sets to verify the effectiveness of the proposed algorithms. |
PDF Full Text Request