Using MLAG in Data Center Network Design

In modern data center network architecture, ensuring high availability, scalability, and efficient resource utilization is crucial. As data center network architecture continue to evolve to meet increasing demands, traditional network solutions often face limitations in bandwidth efficiency and operational complexity.

This article explores how MLAG technology enhances data center network design, improves bandwidth, strengthens network resilience, and simplifies management operations to meet the growing business needs.

MLAG Technology Overview

Multi-Chassis Link Aggregation Group(MLAG) is a cross-device link aggregation mechanism that enables interface bundling across two devices to achieve device-level redundancy. SwitchA1 and SwitchA2 synchronize control plane messages and MAC addresses of LAG interfaces to ensure Layer 2 multicast packets maintain consistent MAC addressing. Downstream devices can be any LACP-capable equipment. When networks are dual-homed via MLAG, connected devices remain unaware that two separate switches are attached at the other end of the links.

The advantages of MLAG include:

High Bandwidth & Reliability:

Delivers scalable bandwidth and enhanced reliability during traffic surges.

Loop-Free Topology with Redundant Backup:

Eliminates complex Spanning Tree Protocol configurations while maintaining redundancy.

Simplified Connectivity:

Supports static LAG or LACP connections to other switches/servers without additional protocols.

Redundancy:

Provides fully active Layer 2 redundancy protection.

Independent Upgrades:

Enables non-disruptive, independent upgrades of either device with minimal service impact.

Legacy Network Designs

In traditional data center architectures without MLAG technology, reliance on Spanning Tree Protocol (STP) for Layer 2 loop prevention causes the blocking of uplink connections, resulting in inefficient bandwidth utilization and a cost-performance imbalance. This issue is particularly evident in the following areas:

Bandwidth waste:

STP blocks redundant links, preventing effective bandwidth from being utilized.

Scalability bottleneck:

As server density increases, the STP-blocked links exacerbate network congestion, particularly in virtualization clusters, where unpredictable latency issues are likely to occur.

Operational complexity:

Even with MSTP for traffic load balancing, administrators must precisely configure VLAN mappings, and configuration errors may lead to suboptimal paths or network disruptions.

Necessity of MLAG in Data Center Network Architecture

In traditional data center network architecture, redundant devices and links are commonly used to enhance resilience and stability. However, this often results in low link utilization and high operational costs. MLAG and stacking technologies can virtualize multiple switches into a single logical switch, simplifying network deployment and reducing management complexity.

Compared to stacking, MLAG offers higher reliability and shorter service interruption time. The diagram below illustrates a typical MLAG topology.

MLAG virtualizes two physical switches into a single logical entity, overcoming the limitations of traditional redundancy designs. Unlike conventional approaches that rely on STP to block redundant links, MLAG enables active-active traffic forwarding, ensuring efficient utilization of all uplinks while preventing loops.

Benefits of MLAG Deployment in Data Center Network Architecture

High Bandwidth & Resiliency:

MLAG enhances north-south and east-west bandwidth through link aggregation and improves network stability with an independent control plane, ensuring that a single point of failure does not impact peer switches. It also eliminates STP blocking, optimizes link utilization, and boosts overall traffic throughput.

Simplified Management:

Eliminates the need for STP configuration, reduces network complexity, and enables unified management of MLAG devices for easier deployment and maintenance. MLAG can prevent Layer 2 loops without requiring Spanning Tree Protocol while enhancing LAG redundancy.

Easy Upgrade:

With independent control planes, software upgrades can be performed without service interruptions, ensuring stable network operation. The two MLAG peer switches can be upgraded independently, with simple operations and low risk, effectively preventing service disruptions during the upgrade process.

MLAG Deployment Schemes

MLAG can be applied to different data center network architecture to eliminate bottlenecks and improve resilience.

3-Tier Data Center Architecture

In traditional 3-tier data center architectures, two MLAG switches at the access or aggregation layer exchange packets via peer links and forward traffic. Since MLAG peer switches are independent devices, they can function as separate OSPF nodes for management and support local traffic prioritization, minimizing east-west traffic. The Dual Active Detection (DAD) feature of MLAG can also be implemented without additional wiring.

Access Layer

: When MLAG is deployed on access switches, servers are dual-homed to two access switches, with both NICs operating in active-active mode. The NICs share the same MAC address and balance traffic based on flow. The connected ports are configured as Eth-Trunk through MLAG, ensuring synchronized MAC addresses and ARP entries.

Aggregation Layer

: MLAG on aggregation switches creates a loop-free logical network between aggregation and access switches, which STP cannot achieve. The system configures the two aggregation switches with a peer link and establishes the downlinks to the same access switch as an Eth-Trunk interface.

2-Tier Spine-Leaf Architecture

MLAG is also applicable to 2-tier spine-leaf architectures. In this architecture, two spine switches are paired as a single MLAG domain, functioning as a single switch connected to the leaf switches. All links in the topology are used for forwarding without blocked ports, and the spine switches can also act as inter-VLAN gateways for the data center.

The figure shows a typical deployment of MLAG in a 2-tier spine-leaf design.

Leaf Layer

: At the leaf layer, servers employ NIC bonding or standard link aggregation to establish multiple active-active connections with leaf switches, ensuring both redundancy and load balancing across all available interfaces.

Spine Layer

: In the spine layer, two data center switches form an MLAG pair, aggregating all uplinks, eliminating blocked ports, and providing interconnect bandwidth across all ports.

MLAG Deployment Practices

In the multi-tenant cloud data center network, the FS solution employs a tiered deployment with 100G PicOS® data center switches at the spine layer and 25G PicOS® data center switches at the leaf layer to deliver:

Node-level Redundancy:

Leaf layer switches form MLAG clusters in pairs, synchronizing the control plane through a Peer-Link and connecting to the spine layer switches via dual uplinks.

In the event of a device or link failure, traffic is automatically redirected to maintain uninterrupted tenant services.

Efficient Resource Sharing:

Physical network resources are shared by multiple tenants, with MLAG providing logical traffic isolation (based on VLAN/VXLAN) to ensure security and QoS.

Automated Scalability:

FS Data Center PicOS® Switch supports the AmpCon-DC management platform for automated operations and configuration, improving efficiency and system reliability.