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. In this paper, we define PLAN, a Policy-Aware virtual machine management scheme to jointly consider DC communication cost reduction through Virtual Machine (VM) migration while meeting network policy requirements.
Introduction: The Challenge of Policy-Blind Migration
In modern cloud computing, Virtual Machine (VM) migration is a fundamental tool for load balancing and resource management. However, this flexibility often overlooks a critical constraint: network policies. Data Center (DC) networks frequently require traffic to traverse a specific sequence of "middleboxes"—such as firewalls (F), load balancers (LB), and intrusion prevention systems (IPS)—to maintain security and performance standards.
When a VM is migrated without considering these service chains, the resulting network paths can become highly inefficient. A "policy-blind" migration might move a VM closer to its destination but significantly further from a required middlebox, leading to increased latency, wasted bandwidth, and potential security violations. Research indicates that ignoring these policies can lead to violations in up to 91% of cases, contributing significantly to data center downtime.
The PLAN Framework
To address these inefficiencies, the researchers introduce PLAN (PoLicy-Aware virtual machine maNagement). PLAN treats VM management as a joint optimization problem that minimizes communication costs while strictly adhering to middlebox traversal sequences.
Core Components of PLAN:
- Cost Modeling: The system defines a communication cost function that weights network links (e.g., core links are prioritized over edge links) and accounts for traffic volume between VMs and their required middleboxes.
- Utility-Based Migration: PLAN calculates a "utility" for potential migrations, defined as the reduction in communication cost minus the overhead of the migration itself (memory state transfer).
- Decentralized Heuristic: Since optimal VM placement is NP-Hard, PLAN employs a decentralized heuristic. Server hypervisors monitor local traffic and trigger migration phases to find optimal destinations based on current policies, avoiding the bottlenecks of a centralized global scheduler.
- Intelligent Initial Placement: PLAN includes algorithms to optimize the initial placement of new VMs, preventing poor configurations before they occur.
Evaluation and Key Results
The PLAN scheme was evaluated using the ns-3 network simulator on a standard fat-tree data center topology. The performance was compared against a "Random Initial Placement" (PLAN-RIP) baseline.
The results demonstrate significant improvements in both efficiency and stability:
- Cost Efficiency: PLAN achieved an average communication cost reduction of 39.06% per migration.
- Rapid Convergence: The system reaches a stable state quickly; 30% of VMs required only a single migration to reach an optimal configuration.
- Minimal Overhead: Less than 1% of VMs required more than two migrations, and none required three or more, highlighting the effectiveness of the heuristic.
Conclusion
PLAN provides a robust framework for aligning elastic cloud resources with rigid network policies. By integrating policy awareness into VM management, data center operators can significantly reduce communication costs and latency while ensuring security compliance. Future work will look into adapting PLAN for highly dynamic policies and exploring its scalability in "mega-scale" data centers.
Summary updated