[Paper Review] Stratos: A Network-Aware Orchestration Layer for Virtual Middleboxes in Clouds

Aaron Gember-Jacobson, Anand Krishnamurthy+7 more|arXiv (Cornell University)|May 1, 2013

Cloud Computing and Resource ManagementComputer Science46 references84 citations

TL;DR

Stratos is a network-aware orchestration layer that enables correct and efficient composition and provisioning of virtual middleboxes in cloud environments using software-defined networking. It ensures forwarding correctness through lightweight, virtualization-aware flow steering and achieves scalable, resource-efficient provisioning via traffic engineering, dynamic flow distribution, and network-aware scaling, reducing resource usage by up to one-third compared to baselines.

ABSTRACT

Enterprises want their in-cloud services to leverage the performance and security benefits that middleboxes offer in traditional deployments. Such virtualized deployments create new opportunities (e.g., flexible scaling) as well as new challenges (e.g., dynamics, multiplexing) for middlebox management tasks such as service composition and provisioning. Unfortunately, enterprises lack systematic tools to efficiently compose and provision in-the-cloud middleboxes and thus fall short of achieving the benefits that cloud-based deployments can offer. To this end, we present the design and implementation of Stratos, an orchestration layer for virtual middleboxes. Stratos provides efficient and correct composition in the presence of dynamic scaling via software-defined networking mechanisms. It ensures efficient and scalable provisioning by combining middlebox-specific traffic engineering, placement, and horizontal scaling strategies. We demonstrate the effectiveness of Stratos using an experimental prototype testbed and large-scale simulations.

Motivation & Objective

• Enterprises seek to leverage the performance and security benefits of middleboxes (e.g., firewalls, IDS) in cloud deployments but lack systematic tools for composing and provisioning them.
• Existing solutions suffer from incorrect forwarding due to middlebox statefulness and packet mangling (e.g., NAT, session termination), especially under dynamic scaling.
• Current provisioning strategies are inefficient and non-scalable, often leading to overprovisioning or poor resource utilization due to network bottlenecks and lack of coordination.
• The paper aims to design and implement Stratos, a novel orchestration layer that ensures correct service chaining and efficient, scalable provisioning of virtual middleboxes in cloud environments.

Proposed method

• Stratos uses a forwarding controller built as a Floodlight module to manage flow steering via lightweight, virtualization-aware SDN mechanisms that avoid steering ambiguity by marginally over-provisioning MBox chains.
• It employs a multi-level provisioning strategy: first using flow distribution to handle transient bottlenecks without prior knowledge of their cause.
• For persistent bottlenecks, it applies network-aware horizontal scaling and VM migration, guided by traffic engineering and placement algorithms to minimize network footprint and maximize resource utilization.
• The resource controller computes optimal flow distributions using linear programming (LP) and performs placement/migration decisions based on inter-rack bandwidth and traffic volume metrics.
• The system decouples flow distribution (non-coordinated) from placement (coordinated), enabling horizontal scaling of controllers via tenant partitioning and modular deployment.
• A prototype implementation (~12K lines of Java) includes a forwarding controller and a standalone resource controller, evaluated via real testbed and large-scale simulations.

Experimental results

Research questions

• RQ1 How can virtual middlebox chains be composed correctly in the presence of packet mangling and dynamic provisioning in cloud environments?
• RQ2 What mechanisms enable efficient and scalable provisioning of virtual middleboxes without prior knowledge of bottleneck types?
• RQ3 To what extent can network-aware flow distribution and placement reduce resource consumption and improve utilization in multi-tenant cloud deployments?
• RQ4 How does Stratos scale under high tenant loads, and can its controllers handle real-time provisioning decisions for hundreds of tenants?

Key findings

• Stratos ensures correct forwarding composition with 0% error rate, outperforming state-of-the-art techniques that exhibit ~19% error rate under mangling and dynamic provisioning.
• The composition mechanism imposes only 1ms overhead per flow per mangling middlebox, enabling low-latency operation.
• Stratos reduces resource usage by up to one-third compared to baseline solutions, achieving higher instance utilization (e.g., 50% of instances process >12Mbps vs. <3Mbps in UniformFlow).
• Network utilization improves significantly: Stratos increases inter-rack traffic by enabling chains to scale closer to demand, with 85% of chains satisfying 30% of demand vs. only 20% in UniformFlow.
• The resource controller sustains 51 flow distribution operations per second and 67 placement operations per second on a single server, supporting hundreds of tenants.
• The forwarding controller can scale using existing SDN techniques, and the system’s control logic is sufficiently responsive to support real-time provisioning under dynamic workloads.