This technology addresses the heavy network load caused by migrating multiple virtual machines simultaneously in data centers. By using cluster-wide deduplication to reduce redundant data transfer, it enables faster migrations, lowers network congestion, and improves overall application performance for datacenter administrators while enhancing resource utilization and reducing operational inefficiencies.
Datacenter clusters often rely on live migration of virtual machines (VMs) to balance workloads and prevent failures. Simultaneous migration of multiple VMs (known as gang migration) in datacenters generates significant network traffic and overloads core network links. This congestion slows down migrations and negatively impacts applications that rely on the network, creating inefficiencies in resource utilization. Traditional migration methods often retransmit duplicate data and take longer to complete, increasing operational costs and reducing overall system efficiency. A more efficient migration approach is needed to reduce network congestion, speed up VM transfers, and improve resource utilization across datacenter clusters.
This invention introduces gang migration using global deduplication (GMGD), a cluster-wide deduplication mechanism that reduces network traffic overhead by identifying identical memory pages across VMs running on different physical machines within a cluster. By leveraging pre-computed hashing and coordinated deduplication servers, the system ensures that only one copy of each identical page is transmitted per target rack, reducing total data transfer. The system is implemented on a QEMU/KVM-based prototype and tested on a 30-node cluster. GMGD reduces network traffic by up to 65% and migration time by up to 42% compared to the standard migration techniques, while also minimizing performance impact on running applications.
![]()
• Eliminates retransmission of identical memory pages across racks using cluster-wide deduplication
• Reduces network traffic by up to 65% and migration time by up to 42%
• Maintains low CPU overhead during deduplication to avoid migration bottlenecks in low-latency LANs
• Scales efficiently in multi-rack clusters, supporting simultaneous migration of multiple VMs
• Integrates with existing virtualization platforms without user-level changes
• Improves overall system efficiency and resource utilization in datacenter environments
• Cloud datacenter VM management for improving resource utilization
• Enterprise server load balancing during peak workloads and maintenance
• High-performance computing cluster migration for memory-intensive simulations
• Edge and telecom datacenter VM relocation for low-latency applications
• United States, 61/992/037, Provisional, Converted, Filed 5/12/2014
• United States, 14/709,957, Patented, Utility, Filed 5/12/2015, Issued 11/21/2017
• United States, 15/818,163, Patented, Continuation, Filed 11/20/2017, Issued 12/18/2018
Prototype - technology validated through implementation and experimental evaluation on a 30-node QEMU/KVM cluster testbed
This technology is available for licensing.
Strong potential for cloud service providers, enterprise datacenters, and infrastructure operators seeking to optimize VM migration efficiency, reduce network congestion, and improve performance in large-scale distributed computing environments.
Related publication:
https://ieeexplore.ieee.org/abstract/document/6546118