What we learned after years of evolution in Data Center architecture, and how you can implement these changes tomorrow.
Data centers have been the backbone of enterprise IT for decades, but traditional network architectures have started to show their limitations. In particular, oversubscription in fabrics, where network devices or links lack sufficient capacity for full-line rate communication, continues to be a persistent challenge. This issue causes bottlenecks, suboptimal forwarding, and affects VM mobility and overall performance.
What most teams do today?
Many organizations still rely on older 3-tier models, relying on spanning tree protocols (STP) to maintain a loop-free topology. This model, despite its long history, suffers from large broadcast domains, limited scalability, and inefficient forwarding paths. As virtualization grows and demands increase, this architecture quickly becomes inadequate, leading to network congestion and unmanageable traffic flows.
Why it fails?
Traditional designs rely heavily on protocols like STP and HSRP (Hot Standby Router Protocol) to manage network traffic, which introduces:
- Suboptimal forwarding: Limited by reliance on traditional routing protocols.
- Scalability issues: Cannot support large-scale virtualization or inter-data center communication.
- Fault tolerance challenges: Spanning Tree Protocol (STP) fails to provide the necessary resilience for modern network requirements.
As network loads increase, these failings become even more apparent, leading to slower speeds and, ultimately, dissatisfaction with the overall performance of data center networks.
Framework / Approach
To address these issues, it’s crucial to shift towards a fabric-based architecture that offers better scalability, reduced latency, and improved overall performance.
Step 1: Define
We must clearly identify the key limitations of the traditional network model, particularly around the areas of oversubscription, inefficient forwarding, and insufficient capacity. Recognizing the limitations of older protocols and understanding modern network requirements, such as VM mobility and seamless inter-DC connections, is key.
Step 2: Diagnose
Evaluate the current architecture of your data center:
- Are you relying on 3-tier designs with STP?
- Is VM mobility causing network congestion or instability?
- Do your uplinks to the spine switches have enough capacity?
Step 3: Decide
Based on the evaluation, choose the appropriate upgrade path. Modernizing to a fabric-based architecture like VXLAN with BGP EVPN, or considering a Clos architecture, can alleviate many of the issues posed by oversubscription and inefficient forwarding.
Step 4: Deliver
- Implement VXLAN for better L3 connectivity and VM mobility across data centers.
- Utilize BGP EVPN for seamless, scalable fabric solutions that integrate with existing networks.
- Focus on upgrading the uplink capacity to spine switches to avoid oversubscription and increase overall throughput.
Case Study / Example
A mid-sized enterprise with an existing 3-tier network design, operating a large number of virtual machines and expanding rapidly. The company was struggling with network bottlenecks and slow data transfers, especially as their workloads became more distributed across multiple data centers.
Actions taken
- Evaluated the current network’s performance and identified oversubscription as the main cause of delays.
- Upgraded the network design to a fabric-based architecture using VXLAN and BGP EVPN.
- Increased uplink capacity between the spine and leaf switches.
- Implemented ECMP for optimal path selection and improved load balancing.
- Deployed the updated network in a phased manner, first in the primary data center, followed by the secondary locations.
Results (Numbers + Timeline)
- Improved throughput: 35% increase in network speed.
- Reduced congestion: Over 40% reduction in traffic bottlenecks.
- Enhanced VM mobility: Seamless inter-data center workload migration.
What Didn’t Work?
While the fabric upgrade solved many problems, the implementation process highlighted that not all network devices were fully compatible with the new VXLAN and EVPN protocols. This required additional adjustments to the hardware and network devices, which delayed full deployment by 3 months.
Playbook / Checklist
- Checklist item 1: Review current architecture to identify network congestion points.
- Checklist item 2: Upgrade to a fabric-based architecture with VXLAN and BGP EVPN.
- Checklist item 3: Ensure uplink capacity is sufficient to prevent oversubscription.
How to start in 30 minutes
- Evaluate your current data center architecture. Identify key bottlenecks and determine if you’re using legacy protocols like STP.
- Learn about VXLAN and BGP EVPN. Understand how these technologies can help modernize your data center fabric.
- Assess hardware compatibility. Check if your existing switches and routers can support the new fabric protocols.
Conclusion & Next Step
The evolution from a traditional 3-tier architecture to a fabric-based network is essential for modern data centers. By eliminating oversubscription and optimizing traffic flows, businesses can enhance performance, scalability, and VM mobility. Start by evaluating your current network, and make a plan for phased upgrades to achieve better, more resilient data center infrastructure.
At TelenceSolutions
We continue to help professionals build scalable, intelligent networks through real-world, hands-on learning — from OSPF and IS-IS fundamentals to BGP, SD-WAN, and AI-driven automation.