The GARR network has an extremely high transmission capacity, up to 200 Gbps for single access, for both download and upload, with a backbone capacity of 23 Tbps. Its Cloud Platform includes five main data centres – two in Rome and one each in Bologna, Bari, and Catania – all interconnected with high-capacity fibre; and with over 100 Points of Presence, it provides advanced services to more than 1,000 sites, ensuring robust and scalable connectivity for around 3 million users across the country.
In 2019, GARR’s data centres relied on a traditional layer 2 fabric architectural model, using Ethernet for transport and VLANs to divide broadcast domains. However, the progressive adoption of virtualisation and containerisation led to a shift in application architecture, resulting in a horizontally distributed model. This shift significantly increased East-West traffic, which met the limitations of the previous model, including suboptimal bandwidth utilization and high convergence times due to the Spanning Tree Protocol (STP).
To create a more efficient and responsive data centre network infrastructure, GARR decided to transition to an IP fabric architectural model based on a Spine-Leaf topology. The new architecture functions as a single layer 3 domain, leveraging routing protocols for improved load balancing, and eliminating STP to reduce convergence times.
Following a competitive tendering process and a Proof-of-Concept, GARR selected Arista Networks, choosing to deploy Arista DCS 7050CX3 and DCS 7050SX3 as spine and leaf switches, respectively, to establish a robust and reliable backbone with high-speed 100 Gbps links. This solution ensures redundancy and high availability by dual-homing leaf switches to spine switches, to create a resilient and efficient data transport network across the data centres.
Key technologies included Arista Extensible Operating System (EOS®), CloudVision®, and Arista Validated Designs (AVD®).
- EOS: Central to Arista’s solutions, EOS simplifies provisioning and maintenance with a single binary software image across all switches, reducing operational complexity.
- CloudVision: This platform enhances automation, providing real-time telemetry, analytics, and cognitive insights through AI and machine learning. It replaces traditional methods with streaming for faster problem resolution and actionable insights.
- AVD: Automating network provisioning and management, AVD ensures consistent configurations, minimizes errors, and streamlines processes across data centres, campuses, and wide-area networks.
Massimo Carboni, GARR deputy director and CTO notes, “We recognized early on that leveraging AVD and CloudVision was a very effective model for network automation, and it has significantly streamlined our operations and enhanced our network’s performance.”
The solution also utilized EVPN/VXLAN technology with BGP as the routing protocol for both underlay and overlay networks. This technology allows for flexible network segmentation and efficient data routing, ensuring the network is future-proof. It also provides the capability to divide into different Virtual Routing and Forwarding (VRF) instances.
The network upgrade resulted in several significant benefits for GARR. The enhanced load-balancing capabilities of the routing protocols ensured even traffic distribution across the network, reducing bottlenecks and improving overall performance. The Spine-Leaf design eliminated the need for STP, significantly reducing convergence times and ensuring a more responsive and reliable network.
The IP Fabric model met the scalability needs of GARR’s data centres, facilitating seamless growth and future expansion. This architecture efficiently supports the increased East-West traffic generated by modern application architectures, ensuring efficient data transport and communication between application components.
Carboni confirms, “The implementation has provided us with the flexibility and adaptability we need to meet our users’ evolving demands while ensuring long-term sustainability. This innovative approach paves the way for new services and applications in our research community”.
