This paper proposes a traffic-aware virtual machine (VM) placement approach to improve the scalability of data center networks. Unlike existing solutions that require changes to network architecture and routing protocols, this method optimizes VM placement on host machines to align traffic patterns with communication distances between VMs. VMs with high mutual bandwidth usage are placed on hosts close to each other, reducing network costs. The VM placement problem is formulated as an optimization problem and proven to be NP-hard. A two-tier approximate algorithm is designed to efficiently solve the problem for large sizes. The paper also analyzes the impact of traffic patterns and network architectures on the performance gains of traffic-aware VM placement. Using traffic traces from production data centers, the proposed algorithm is shown to significantly improve performance compared to existing generic methods. The study shows that multi-level network architectures, such as BCube, benefit the most from traffic-aware VM placement, while architectures designed for uniform traffic patterns, such as VL2, see little benefit. The paper also presents a two-tier algorithm called Cluster-and-Cut, which partitions VMs and hosts into clusters and matches them at cluster and individual levels. The algorithm is evaluated using production data center traces and shows significant improvements in aggregate traffic reduction and computational efficiency. The analysis also shows that the potential benefit of optimizing TVMPP depends on network architecture and traffic patterns. The paper concludes that traffic-aware VM placement can significantly improve network scalability, especially in data centers with heterogeneous traffic patterns.This paper proposes a traffic-aware virtual machine (VM) placement approach to improve the scalability of data center networks. Unlike existing solutions that require changes to network architecture and routing protocols, this method optimizes VM placement on host machines to align traffic patterns with communication distances between VMs. VMs with high mutual bandwidth usage are placed on hosts close to each other, reducing network costs. The VM placement problem is formulated as an optimization problem and proven to be NP-hard. A two-tier approximate algorithm is designed to efficiently solve the problem for large sizes. The paper also analyzes the impact of traffic patterns and network architectures on the performance gains of traffic-aware VM placement. Using traffic traces from production data centers, the proposed algorithm is shown to significantly improve performance compared to existing generic methods. The study shows that multi-level network architectures, such as BCube, benefit the most from traffic-aware VM placement, while architectures designed for uniform traffic patterns, such as VL2, see little benefit. The paper also presents a two-tier algorithm called Cluster-and-Cut, which partitions VMs and hosts into clusters and matches them at cluster and individual levels. The algorithm is evaluated using production data center traces and shows significant improvements in aggregate traffic reduction and computational efficiency. The analysis also shows that the potential benefit of optimizing TVMPP depends on network architecture and traffic patterns. The paper concludes that traffic-aware VM placement can significantly improve network scalability, especially in data centers with heterogeneous traffic patterns.