This paper addresses the scalability issue in modern data center networks by proposing a traffic-aware virtual machine (VM) placement approach. The authors formulate the Traffic-aware VM Placement Problem (TVMPP) as an optimization problem, where the goal is to minimize the aggregate traffic rates perceived by each switch by optimizing the placement of VMs on host machines. They prove that TVMPP is NP-hard and propose a two-tier approximate algorithm, Cluster-and-Cut, to efficiently solve it for large problem sizes. The algorithm first partitions VMs and hosts into clusters separately, then matches VMs and hosts at the cluster level and subsequently at the individual level. The effectiveness of the proposed algorithm is evaluated using traffic traces from production data centers, showing significant performance improvements compared to existing methods. The paper also analyzes the impact of different network architectures and traffic patterns on the potential performance gains from optimal VM placement, finding that multi-level network architectures like BCube benefit more from optimal VM placement than uniform traffic pattern architectures like VL2. The results highlight the potential benefits of optimizing VM placement for improving network scalability, especially in data centers with heterogeneous traffic patterns.This paper addresses the scalability issue in modern data center networks by proposing a traffic-aware virtual machine (VM) placement approach. The authors formulate the Traffic-aware VM Placement Problem (TVMPP) as an optimization problem, where the goal is to minimize the aggregate traffic rates perceived by each switch by optimizing the placement of VMs on host machines. They prove that TVMPP is NP-hard and propose a two-tier approximate algorithm, Cluster-and-Cut, to efficiently solve it for large problem sizes. The algorithm first partitions VMs and hosts into clusters separately, then matches VMs and hosts at the cluster level and subsequently at the individual level. The effectiveness of the proposed algorithm is evaluated using traffic traces from production data centers, showing significant performance improvements compared to existing methods. The paper also analyzes the impact of different network architectures and traffic patterns on the potential performance gains from optimal VM placement, finding that multi-level network architectures like BCube benefit more from optimal VM placement than uniform traffic pattern architectures like VL2. The results highlight the potential benefits of optimizing VM placement for improving network scalability, especially in data centers with heterogeneous traffic patterns.