Enabled by virtualization technologies, various multi-tier applications are hosted by virtual machines (VMs) in cloud data centers. Live migration of multi-tier applications across geographically distributed data centers is important for load management, power saving, routine server maintenance and quality-of-service. Different from a single-VM migration, VMs in a multi-tier application are closely correlated, which results in a correlated VM migrations problem. Current live migration algorithms for single-VM cause significant application performance degradation because intermediate data exchange between different VMs suffers relatively low bandwidth and high latency across distributed data centers. A coordination system called VMbuddies is designed for correlated VM migrations in the cloud. Particularly, an adaptive network bandwidth allocation algorithm is proposed to minimize the migration cost in terms of migration completion time, network traffic and migration downtime. Experiments using a public benchmark show that VMbuddies significantly reduces the performance degradation and migration cost of multi-tier applications.