Case studyBanking & Financial Services

Building a Tier III-ready software-defined datacenter for one of Ethiopia’s largest banks

A software-defined datacenter network and security core, engineered to Tier III principles — delivered on schedule, formally recognized by the bank’s CEO and CIO, and still the foundation of the bank’s digital platform today.

Delivered2022project completed
ClientOne of Ethiopia’s largest private banks
Solution areasNetworking & Connectivity · Cybersecurity · Data Center, Cloud & AI Infrastructure
TimelineNine months, delivered on schedule
01 — Challenge

The challenge

Ethiopian banking was changing fast. Mobile money, online channels, agent banking, and a wave of new digital services were arriving on top of core banking systems that had been designed for a smaller, slower institution. For one of the country’s largest private banks, the constraint was no longer software or ambition — it was the datacenter beneath everything.

The existing network had grown organically over years: a traditional three-tier design with layer-2 domains stretched across the datacenter, manual configuration on every device, and security concentrated at the perimeter with little control over traffic moving inside the datacenter. That combination created three problems the bank could no longer accept:

01Availability risk Spanning-tree topologies and hand-configured devices meant that a single change, or a single failure, could ripple across services. For a bank, unplanned downtime is measured in reputational cost, not minutes.
02No agility Bringing a new application or environment online meant days of manual network work. The business wanted hours.
03East-west blind spot Once traffic entered the datacenter, it moved between servers largely uninspected. Any compromise of one system could become a compromise of many.

The bank set a demanding target: a Tier III-ready datacenter — redundant, concurrently maintainable, with no single point of failure — where the network fabric and the security core were designed as one system, not bolted together afterward. And it had to be delivered without disrupting live banking operations.

02 — Approach

Our approach

We treated this as an architecture engagement first and a deployment second.

  1. 01

    Discovery

    Before proposing a design, we mapped the bank’s application landscape: which systems talked to which, over what protocols, at what volumes, with what latency sensitivity. We captured the existing security policy and traffic flows, identified the crown-jewel systems that warranted the tightest segmentation, and modeled the bank’s growth plans — because a datacenter is sized for the workloads it will carry in five years, not the ones it carries today.

  2. 02

    Design review with the bank’s teams

    The architecture was reviewed jointly with the bank’s infrastructure and cybersecurity leadership against the standards a financial institution answers to — regulatory expectations for availability and control, and international good practice for datacenter design. Every zone, every policy boundary, every redundancy path was agreed before procurement began.

  3. 03

    Phased, parallel-run delivery

    The new fabric was built alongside the existing network. Services were migrated in agreed maintenance windows, one workload group at a time, with the old path retained until each migration was verified in production. Rollback was always available. At no point was the bank asked to accept a “big-bang” cutover.

  4. 04

    Test, accept, transfer

    Each phase closed with documented testing — failover, throughput, policy enforcement — and formal acceptance by the bank. The engagement ended with structured knowledge transfer so the bank’s own engineers could operate, extend, and troubleshoot the platform.

03 — Solution

The solution

Software-defined datacenter fabric

The heart of the new datacenter is a leaf-spine fabric built on EVPN/VXLAN — the architecture that has replaced the fragile three-tier design across modern datacenters worldwide.

  • Any workload, any rack, one hop. Every leaf connects to every spine, so traffic between any two servers crosses a single, predictable path. There is no spanning tree in the fabric, and no blocked links wasting capacity.
  • Layer-2 mobility without layer-2 risk. VXLAN overlays give applications the flat, mobile connectivity they expect, while the underlying network stays a clean, routed, loop-free fabric.
  • Growth by adding leaves. New capacity means racking a new leaf pair — not redesigning the core.
  • Redundancy everywhere. Dual spines, dual uplinks from every leaf, and dual-homed servers, so any single component can fail — or be taken down for maintenance — without a service interruption. This is what “Tier III-ready” means in practice.

Datacenter security core

Security was engineered into the fabric, not appended to its edge.

  • Next-generation firewalls at the datacenter boundary and between security zones, enforcing policy based on the application and the identity behind a connection rather than IP addresses alone.
  • East-west inspection. Traffic between zones — for example, between the channel-facing tier and the core-banking tier — now passes through inspection, so lateral movement from a compromised system is stopped at the zone boundary.
  • Segmentation aligned to the business. Zones were defined from the dependency map: what must talk to what, and nothing more.
  • High-availability firewall clustering, so security enforcement never becomes the single point of failure it was designed to eliminate.

Automation and observability from day one

  • Centralized fabric management — configuration lives as intended state in a single system, changes are made once and pushed consistently, and drift is visible immediately rather than discovered during an outage.
  • Streaming telemetry from every switch feeds live visibility of the fabric’s health and traffic, turning the network from a black box into an instrumented platform.
04 — Result

The result

The datacenter was delivered on schedule and formally accepted. It became — and remains — the platform beneath the bank’s digital services.

Teamwork, proficiency, and diligence
From the client’s letter

The bank’s Chief Executive Officer and Chief Information Officer jointly issued a certificate of appreciation to eSafe for the “outstanding contribution” to the project and the “teamwork, proficiency, and diligence” shown throughout — recognition that came from the top of the institution, not only from its IT department.

The clearest measure, though, is what happened next. In the years since go-live, the bank has entrusted eSafe with the platform’s entire lifecycle: multi-year license and support renewals for the fabric and for the firewall estate, expansion optics as the fabric grew, an enterprise email-security program covering several thousand mailboxes, and — in 2025 — a new campus software-defined network for its head office, built on the same principles and awarded to eSafe through open competition. A datacenter is a decade-long commitment; the bank has made it with eSafe.

05 — Why it worked

Why it worked

Design from dependencies, not defaults

The fabric served the bank’s applications because we mapped them before we drew a diagram.

Security as a property of the architecture

East-west inspection and zone-based segmentation were in the design from the first review — not a retrofit.

Parallel-run, never big-bang

A bank cannot afford a maintenance window that goes wrong. Every cutover was reversible by design.

Leave the knowledge behind

The bank’s own engineers operate the platform today. That was a deliverable, not an afterthought.

Client identity withheld under our confidentiality commitments. Details, quotations, and outcomes are drawn from the client’s formal project acceptance and recommendation letters, held on file.

More case studies

Other engagements, same discipline

Next step

Have a project like this one?

Tell us your infrastructure reality and your constraints — we’ll show you how we’d approach it.

Talk to a pro