<p>Dynamic Virtual Machine Consolidation (DVMC) is a key mechanism for developing an energy-aware dynamic resource management in cloud datacenter. The basic idea is to balance the hosts' load by migrating Virtual Machines (VMs) from overloaded and underloaded host to normal host and turning the underloaded host into sleep mode. Each VM migration leads performance degradation and Service Level Agreement Violation (SLAV). It is necessary to enhance performance while dealing with energy–SLAV tradeoff in DVMC. Therefore, this paper proposes an improved DVMC model named RLSK_US which consists four phases: 1) in first phase, Robust Logistic Regression algorithm is proposed to detect overloaded host. This algorithm utilizes regression and adaptive approaches both; 2) second phase proposes an SLA algorithm to detect underloaded host by incorporating the NVM of the host with CPU utilization; 3) third phase proposes a Knapsack based VM selection algorithm that selects VM having higher ratio of CPU utilization to VM migration time; 4) fourth phase proposes Utilization-SLA aware VM placement algorithm that allocates migrated VMs to appropriate host by selecting host having higher correlation factor with VM. The proposed RLSK_US is evaluated using real workload traces in CloudSim and compared with other existing benchmark algorithms. Simulation results prove that proposed model outperforms others. It improves SLAV by 77% and ESV (product of Energy and SLAV) by 83% compared to the best competitive algorithm.</p>