PLAN: A Policy-Aware VM Management Scheme for Cloud Data Centres

Weijia Jia, Lin Cui, Fung Po Tso, Dimitrios P. Pezaros, Weijia Jia 0001

TL;DR

PLAN (PoLicy-Aware and Network-aware VM management) is a scheme designed to optimize Virtual Machine (VM) placement in cloud data centers by jointly considering server resources, communication costs, and strict network security policies. By accounting for the specific "middleboxes" (like firewalls and load balancers) that traffic must traverse, PLAN reduces network-wide communication costs by up to 38% while ensuring no policy violations occur during migration.

What Problem Does This Paper Address?

In modern multi-tenant cloud data centers, network traffic doesn't just go from Point A to Point B; it must often pass through a "service chain" of middleboxes—firewalls, Intrusion Detection Systems (IDS), and load balancers—to satisfy security and performance policies.

The challenge arises during VM migration. Standard management tools move VMs to consolidate server usage and save power, but they often ignore these network policies. If a VM moves to a server far away from its required middlebox, the traffic path becomes long and inefficient (a problem known as "tromboning"), wasting bandwidth and increasing latency. Worse, if the migration isn't handled carefully, traffic might bypass a security middlebox entirely, leading to policy violations.

Methods

The authors formulate the PLAN problem as an optimization task: find a new VM allocation that maximizes "utility" (the reduction in communication cost minus the overhead of the migration itself) while respecting server capacity and policy constraints.

Key components of their approach include:

  • Mathematical Modeling: They define communication cost based on the entire path, including mandatory middlebox hops, weighted by the "cost" of the network layer (e.g., core links are more "expensive" than edge links).
  • Complexity Analysis: They prove the problem is NP-Hard by reducing it to the Multiple Knapsack Problem.
  • Decentralized Heuristic Algorithm: Since a centralized solution is infeasible for massive data centers, they propose a decentralized greedy algorithm.
  • PLAN-VM: Hypervisors periodically check if moving a local VM to a different server would increase total utility.
  • PLAN-Server: Destination servers accept or reject migration requests based on their remaining CPU/RAM resources.
  • Hybrid Implementation: The system uses a partly decentralized architecture. Computationally expensive cost calculations happen at the hypervisor level (using Xen's Dom0), while a centralized SDN controller provides the global network topology and flow statistics.

Datasets

The authors evaluated PLAN using the ns-3 network simulator and a physical testbed:

  • Simulation Environment: A large-scale Fat-tree topology with 250 servers and 2,320 VMs.
  • Traffic Profile: Each VM averaged 10 random outgoing connections (CBR traffic).
  • Policy Configuration: 90% of flows were constrained by policies requiring them to traverse 1–3 middleboxes (Firewall, IPS, or Load Balancer).
  • Testbed: A physical setup using Xen hypervisors and Open vSwitch to measure the performance overhead of the flow-tracking modules.

Key Insights and Ideas

  • Policy-Awareness vs. Network-Awareness: Existing "network-aware" schemes (like S-CORE) cluster VMs that talk to each other. However, PLAN shows that if you don't account for the middleboxes between those VMs, you still end up with sub-optimal paths.
  • Initial Placement is Half the Battle: The authors found that using a policy-aware algorithm for the initial placement of VMs significantly reduces the need for migrations later, lowering the overall system overhead.
  • Adaptability to Dynamic Policies: PLAN can react to real-time changes in network configuration. If a security policy is disabled (making a flow "policy-free"), PLAN quickly identifies that the VM can be moved to a more efficient location to localize traffic.
  • Scalability through Decentralization: By offloading the "decision-making" to individual hypervisors, the system avoids the bottleneck of a single central controller trying to manage millions of flows.

Key Results

  • Cost Reduction: PLAN reduced topology-wide communication costs by 38% compared to non-policy-aware schemes.
  • Path Efficiency: The average end-to-end route length was reduced by approximately 20%.
  • Congestion Mitigation: PLAN reduced link utilization at the core and aggregation layers by 30.5% and 7%, respectively, effectively creating more "headroom" in the network.
  • Convergence: The decentralized algorithm is stable; most VMs only needed to migrate once to reach an optimized state.
  • Low Overhead: Testbed results showed that the flow-tracking module consumed less than 5% CPU utilization even when managing 10,000 concurrent flows.

Limitations and Future Work

  • Fixed Middlebox Locations: The current model assumes middleboxes are at fixed locations. In a fully virtualized environment (NFV), middleboxes themselves could be migrated, which adds another layer of complexity to the optimization.
  • Traffic Predictability: The algorithm relies on current flow statistics. Integrating traffic prediction models could help PLAN make more proactive migration decisions before congestion occurs.
  • Migration Interference: While the paper accounts for the cost of a single migration, it doesn't deeply explore the "network interference" caused when hundreds of VMs migrate simultaneously.

Summary updated

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 violations, performance degradation and security 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% over diverse aggregate traffic and configuration policies.
BibTeX
@inproceedings{jia2015plan,
  author = {Weijia Jia and Lin Cui and Fung Po Tso and Dimitrios P. Pezaros and Weijia Jia},
  title = {{PLAN: A Policy-Aware VM Management Scheme for Cloud Data Centres}},
  booktitle = {Ucc},
  year = {2015},
}

Loading related content...

Figure