Mosaic: Policy Homomorphic Network Extension

Li Erran Li, Michael F. Nowlan, Y. R. Yang

Abstract

As enterprises migrate workloads to hybrid cloud environments, maintaining consistent security and performance policies becomes a significant challenge. Traditional Layer 2 (L2) network extensions often restore connectivity but bypass critical middle-boxes like firewalls and intrusion prevention systems (IPS). This paper introduces Mosaic, a framework that ensures "policy homomorphism"—the preservation of logical network behavior across physical topology changes. By abstracting policies into way-points and scopes, Mosaic enables seamless server migration while strictly enforcing traversal through necessary security and performance nodes.

Introduction: The Challenge of Hybrid Cloud Migration

The shift toward private and public clouds (e.g., Amazon VPC) introduces a disconnect between physical connectivity and logical policy. In a traditional local data center, the physical layout often dictates that traffic must pass through specific security appliances. However, when a server is moved to a remote cloud, a simple L2 extension might allow traffic to bypass these "middle-boxes," leading to security policy violations, broken TCP sessions due to stateful load balancer bypass, and reachability issues.

A typical enterprise network environment with various middle-boxes and cloud extensions.

The core problem is that connectivity does not equal policy compliance. A successful migration must ensure that the network's logical behavior remains identical, a property the authors term policy homomorphism.

The Mosaic Framework

Mosaic addresses these challenges through a two-stage approach: policy specification and transformation primitives.

Policy Specification

Policies are defined using two primary abstractions:

  1. Way-points: Specific nodes (e.g., firewalls, IPS) and their configurations that a packet must traverse in a defined sequence.
  2. Scopes: The boundaries of a security zone, defining the maximum set of nodes a packet is permitted to reach.

Transformation Primitives

To implement these policies in a hybrid environment, Mosaic utilizes three key primitives:

  • Mosaic Proxy: This "least-disruption" method uses a proxy switch ($S_{proxy}$) to "hairpin" traffic from the remote data center back through the original local policy boxes. This allows the reuse of existing, expensive hardware configurations.
  • Mosaic Mirror: This replicates critical policy boxes in the remote data center to minimize latency.
  • Mosaic Policy Relocation: This optimizes the path by moving software-based policy contexts (like virtual firewall rules) to existing devices along the new network path.
Illustration of traffic paths (Path1 and Path2) between an enterprise intranet and a private data center via an L2 extension.

Notably, Mosaic is designed to work with legacy hardware, avoiding the need for specialized technologies like OpenFlow.

Evaluation and Results

The framework was evaluated using configuration data from a large-scale campus network comprising over 50 routers and 1,000 switches. The study revealed that a naive migration strategy resulted in up to 80% of network paths violating security policies, particularly for external traffic.

In contrast, Mosaic successfully eliminated all policy violations by correctly routing traffic through required way-points. While the "Proxy" approach introduces some additional latency due to the "hairpinning" effect, the evaluation showed that the increase in average path length was manageable and often a worthwhile trade-off for the simplicity of reusing existing security infrastructure.

Conclusion

Mosaic provides a formal and practical foundation for extending enterprise networks to the cloud without compromising security or performance. By focusing on policy homomorphism rather than just connectivity, it allows administrators to leverage cloud scalability while maintaining the integrity of their internal network rules. Future work will look toward automating the optimization between proxying and mirroring and adapting to the dynamic policy requirements of modern threat environments.

Summary updated

BibTeX
@inproceedings{li2010mosaic,
  author = {Li Erran Li and Michael F. Nowlan and Y. R. Yang},
  title = {{Mosaic: Policy Homomorphic Network Extension}},
  booktitle = {LADIS},
  year = {2010},
  doi = {10.1145/1859184.1859195},
}
Figure