Abstract
Policies play an important role in network configuration and therefore in offering secure and high performance services especially over multi-tenant Cloud Data Center (DC) environments. At the same time, elastic resource provisioning through virtualization often disregards policy requirements, assuming that the policy implementation is handled by the underlying network infrastructure. This can result in policy vulnerabilities. In this paper, we define PLAN, a PoLicy-Aware and Network-aware VM management scheme to jointly consider DC communication cost reduction through Virtual Machine (VM) migration while meeting network policy requirements. We show that the problem is NP-hard and derive an efficient approximate algorithm to reduce communication cost while adhering to policy constraints. Through extensive evaluation, we show that PLAN can reduce topology-wide communication cost by 38 percent over diverse aggregate traffic and configuration policies.
Introduction: The Policy-Blind Migration Problem
In modern cloud data centers, Virtual Machine (VM) migration is a cornerstone of resource management, allowing for dynamic consolidation and power savings. However, traditional migration strategies are often "policy-blind," focusing primarily on server-side metrics like CPU and RAM while ignoring the underlying network policies. These policies frequently mandate that traffic traverse specific "middleboxes"—such as firewalls, load balancers, and intrusion detection systems (IDS)—in a strict sequence.
When VMs are moved without considering these requirements, the result is often a "zig-zag" traffic pattern. Packets must travel across the data center to reach a required middlebox and then double back to the destination VM. This not only increases latency but also significantly inflates "east-west" traffic, leading to network congestion and potential security vulnerabilities if policies are bypassed. Research indicates that policy-blind placement can lead to policy violations in up to 91% of cases.
The PLAN Framework: Decentralized Optimization
To address these challenges, the authors propose PLAN (PoLicy-Aware and Network-aware), a management scheme that treats network policies as primary constraints for VM placement. The core objective is to minimize total communication costs while strictly adhering to middlebox traversal sequences.
The authors prove that this joint optimization problem is NP-hard. To solve it efficiently at scale, PLAN employs a decentralized heuristic approach consisting of two main routines:
- PLAN-VM: Operates on source hypervisors to monitor traffic patterns and calculate the "utility" of potential migrations. Utility is defined as the reduction in communication cost minus the migration overhead.
- PLAN-Server: Operates on potential destination servers to manage local resource availability and accept or reject migration requests based on local capacity.
By decentralizing the decision-making process, PLAN avoids the bottlenecks associated with centralized SDN controllers in massive data center environments, ensuring scalability for production use.
Performance and Scalability
The system was evaluated using extensive ns-3 simulations on a 250-server fat-tree topology and a physical Xen-based testbed. The results demonstrate that PLAN significantly outperforms policy-blind alternatives like S-CORE.
Key findings include:
- Communication Cost Reduction: PLAN reduced topology-wide communication costs by 38% compared to state-of-the-art policy-blind schemes.
- Route Efficiency: Average end-to-end route lengths were reduced by 10-20%, effectively localizing traffic at the network edge.
- Migration Stability: By incorporating a policy-aware initial placement strategy, PLAN reduces the need for subsequent migrations. Only 30% of VMs required migration to reach an optimized state, compared to 60% in traditional systems.
Conclusion
PLAN represents a significant shift in cloud resource management by integrating network policy awareness directly into the VM placement logic. By localizing traffic near required middleboxes and using a scalable, decentralized architecture, PLAN ensures that security and performance policies are met without sacrificing network efficiency. Future work may extend this framework to include the joint migration of Virtual Network Functions (VNFs) alongside VMs for even greater optimization.
Summary updated