[Paper Review] LazyCtrl: Scalable Network Control for Cloud Data Centers

Kai Zheng, Lin Wang+4 more|arXiv (Cornell University)|Apr 10, 2015

Software-Defined Networks and 5G15 references3 citations

TL;DR

LazyCtrl proposes a hybrid network control plane for cloud data centers that reduces central controller load by grouping edge switches based on traffic locality, enabling distributed control within groups while offloading only rare inter-group events to the controller. This approach reduces controller workload by up to 82% and improves forwarding latency, significantly enhancing scalability over fully centralized SDN designs.

ABSTRACT

The advent of software defined networking enables flexible, reliable and feature-rich control planes for data center networks. However, the tight coupling of centralized control and complete visibility leads to a wide range of issues among which scalability has risen to prominence. To address this, we present LazyCtrl, a novel hybrid control plane design for data center networks where network control is carried out by distributed control mechanisms inside independent groups of switches while complemented with a global controller. Our design is motivated by the observation that data center traffic is usually highly skewed and thus edge switches can be grouped according to traffic locality. LazyCtrl aims at bringing laziness to the global controller by dynamically devolving most of the control tasks to independent switch groups to process frequent intra-group events near datapaths while handling rare inter-group or other specified events by the controller. We implement LazyCtrl and build a prototype based on Open vSwich and Floodlight. Trace-driven experiments on our prototype show that an effective switch grouping is easy to maintain in multi-tenant clouds and the central controller can be significantly shielded by staying lazy, with its workload reduced by up to 82%.

Motivation & Objective

• To address the scalability limitations of fully centralized SDN controllers in large-scale data centers.
• To reduce the controller's workload by minimizing its involvement in frequent intra-group traffic events.
• To maintain low forwarding latency while decoupling centralized control from complete network visibility.
• To enable practical, dynamic switch grouping that adapts to changing traffic patterns in multi-tenant environments.
• To achieve high compatibility with existing OpenFlow-based SDN solutions while improving system scalability.

Proposed method

• Dynamically group edge switches into local control groups based on traffic communication affinity to exploit traffic locality.
• Delegate coarse-grained control of intra-group traffic to distributed control mechanisms within each switch group, reducing controller involvement.
• Use a Bloom Filter-based Global Forwarding Information Base (G-FIB) to efficiently manage inter-group forwarding with low memory overhead.
• Apply an incremental update mechanism (IncUpdate) to maintain effective switch groupings with minimal controller overhead.
• Integrate with Open vSwitch and Floodlight to implement a fully functional prototype for evaluation.
• Leverage the 'lazy' principle: the controller remains idle for frequent local events, only intervening for rare or fine-grained inter-group events.

Experimental results

Research questions

• RQ1 Can traffic locality in data centers be exploited to reduce the workload of a centralized SDN controller?
• RQ2 How effective is dynamic switch grouping in maintaining controller laziness under real-world traffic patterns?
• RQ3 To what extent can distributed control within switch groups reduce controller load and improve forwarding latency?
• RQ4 What is the practical cost of maintaining dynamic switch groupings in terms of update frequency and controller overhead?
• RQ5 How does the hybrid control design compare to fully centralized SDN in terms of scalability and performance?

Key findings

• LazyCtrl reduces the central controller's workload by up to 82% in trace-driven experiments, demonstrating significant scalability improvements.
• Controller workload remains stable during peak hours on real traffic traces, as most traffic is intra-group and transparent to the controller.
• The incremental update function increases update frequency only moderately (up to 34 updates per hour on expanded traces), indicating feasible maintenance of effective groupings.
• Cold-cache forwarding latency for intra-group traffic is reduced by over 90% (0.83 ms vs. 15.06 ms) compared to standard OpenFlow, due to local forwarding without controller intervention.
• Steady-state forwarding latency is reduced by approximately 10% on average, attributed to lower controller load and faster topology learning.
• The Bloom Filter-based G-FIB achieves a false positive rate below 0.1% with only 92,160 bytes of memory per switch for a 46-switch group, ensuring low storage overhead.