Data Center
Data center networking
Design, build and migration of data center networks: spine-leaf fabrics, EVPN-VXLAN overlays, 25/100/400G upgrades, primary-to-DR interconnect and GPU cluster networking, from enterprise server rooms to colocation cages across India.
Request Proposal WhatsApp NowWhat we design and build
- Spine-leaf fabrics: Replacing ageing three-tier core/aggregation/access designs with a two-tier Clos fabric. Every server is the same number of hops from every other, so east-west traffic between virtualised and containerised workloads stops bottlenecking at the core.
- EVPN-VXLAN overlays: Layer 2 extension where applications still need it, VRF-based segmentation between environments (production, UAT, DMZ, management), and distributed anycast gateways, so a VM keeps its gateway when it moves.
- Speed upgrades: Moving server access from 10G to 25G and spines from 40G to 100G, or to 400G where the traffic justifies it, with an optics and cabling plan that reuses existing fibre wherever possible.
- Storage networking: Lossless Ethernet (PFC/ECN) for iSCSI and NVMe over TCP or RoCE, or Fibre Channel where you already have it. See storage and SAN/NAS.
- Data center interconnect (DCI): Primary-to-DR links over dark fibre, DWDM or carrier circuits, with EVPN multi-site where workloads need to move between sites. Many Indian enterprises run their primary site in Mumbai or Navi Mumbai and DR in Chennai, Hyderabad or Bengaluru, and BFSI regulators expect DR switch-over to be tested, not just documented.
- GPU and AI cluster networking: A separate high-bandwidth, low-latency back-end fabric for training clusters (rail-optimised, lossless RoCEv2) alongside the normal front-end network. See high-performance computing.
- Out-of-band management and automation: An independent console network (see Opengear) so a bad change never locks you out, plus NetBox as the source of truth and Ansible-driven configuration. See DCIM and automation.
Platforms we work with
Cisco Nexus (NX-OS and ACI), Arista EOS with CloudVision, Juniper QFX with Apstra, and HPE Aruba CX. We recommend the platform that fits your team's skills and existing tooling, not just the box with the best datasheet. A fabric that the in-house team cannot operate at 2 a.m. is the wrong fabric.
Typical projects
- Three-tier to spine-leaf migration in a live data center: The new fabric is built alongside the old one, VXLAN gateways bridge existing VLANs during cutover, and racks move in planned windows, each with a tested rollback.
- End-of-life refresh: Retiring end-of-support switches (for example, older Nexus 5000/7000 estates) before they become an audit finding. See EOL/EOS management and Cisco lifecycle upgrades.
- Move to colocation: Designing the network for a new colocation cage, including cross-connects, carrier diversity and a staged migration from an on-premises server room.
- Hybrid cloud edge: Connecting the fabric to AWS, Azure or GCP through dedicated interconnects with consistent segmentation on both sides.
What you receive
A high-level and low-level design, IP/VLAN/VRF and routing plans, a rack-by-rack cabling and optics bill of materials, a migration runbook with rollback steps, a test plan with signed acceptance results, and as-built documentation loaded into NetBox. See network design and documentation. After go-live, the fabric can be monitored by our NOC or handed over to your team with knowledge-transfer sessions.
FAQ
- Should we choose Cisco ACI or a standards-based EVPN-VXLAN fabric?
- ACI provides a policy-driven controller model and suits teams that are already invested in Cisco tooling. Standards-based EVPN-VXLAN, with or without a controller such as CloudVision or Apstra, is more portable across vendors. We compare both against your team's skills, automation plans and renewal costs before recommending one.
- Can a data center network be migrated without downtime?
- Usually, it can be done without an outage that users notice. Building the new fabric in parallel and bridging VLANs during the move lets most workloads migrate in short maintenance windows. Truly zero-downtime cutovers depend on applications being redundant across hosts, which we check during design.
- Do you work inside third-party colocation facilities?
- Yes. We handle the remote-hands coordination, cross-connect orders and access procedures for the major colocation providers.
- How big does a data center need to be for spine-leaf?
- There is no minimum size. Even a pair of spines and four leaves gives a small server room predictable performance and room to grow without a redesign.
See also: all data center services, core and distribution switching.
Still need assistance?
Book Free ConsultationExplore all Data Center services
Get in touch
Request a Callback
Drop your details and we'll call you back within one business day — or reach us directly on +91 81151 74447.