Retail refresh: MPLS exit and cloud-first store traffic
Northline Retail operates 320 stores and two data centers. Its MPLS contract expires in nine months and finance has frozen the WAN budget at today’s spend. Roughly 60% of applications are now SaaS, but POS authorization, pharmacy dispensing, and store back-office ERP remain in the data centers. Each store peaks at about 40 Mbps during holiday promotions; average is 8 Mbps. The CIO wants dual internet at every store with LTE backup for POS, guest Wi-Fi separated from store operations, and PCI-DSS scope reduced. Store sizes range from 2,000 to 18,000 square feet, and 70 stores are in malls with shared landlord fiber. IT has three network engineers, no overnight staff, and no current per-application traffic baseline. Uptime target is 99.95% for POS. Before selecting hardware, the design team must convert this into testable requirements.
1. Identify the missing information you must gather before choosing an SD-WAN architecture, and write the discovery questions you would ask business and store-operations stakeholders. Explain why each matters.
Read the worked answer
Missing information falls into four buckets:
- Traffic and application behavior - per-application flow matrix (source/destination, ports, latency/jitter/loss tolerance), east-west store traffic such as cameras to NVR, and peak concurrent sessions. Without this you cannot size LTE, choose local breakout, or set SLA thresholds.
- PCI scope - where cardholder data is stored and transmitted, whether the store is part of the CDE, tokenization/P2PE status, and required segmentation controls. This determines whether guest, POS, and cameras can share any device or VLAN.
- Site realities - closet space and power, landlord fiber rules, whether a second carrier is allowed, LTE signal strength and monthly cap, and static IP availability. CGNAT and shared landlord fiber directly change the overlay design.
- Operations - who dispatches LTE, acceptable change windows, monitoring ownership, and whether three engineers can run BGP or need a controller.
Useful discovery questions: Which applications must stay on-premises after the cloud migration? What is the dollar loss per minute of POS downtime? Does the mall allow a second physical entry point? What LTE cap is acceptable before throttling? Who tests failover today, and how often? Which stores have only one power circuit? What PCI evidence does the assessor expect for store segmentation?
Each question ties to a design decision: traffic matrix to link sizing, PCI to zoning, site realities to redundancy, and operations to automation and routing complexity.
2. Classify the known statements into functional requirements, non-functional requirements, and constraints. Then state two design principles that follow and one requirement you would push back on.
Read the worked answer
Functional requirements describe behavior: dual internet at each store, LTE backup for POS, guest Wi-Fi separation from store operations, and reduced PCI scope. These are capabilities the design must deliver.
Non-functional requirements describe quality: 99.95% POS availability, 40 Mbps peak per store, acceptable POS transaction latency, and support for 320 stores. They must be testable, so convert them into measurable SLAs such as POS round trip under 150 ms and failover under one second.
Constraints limit the solution: flat WAN budget, three engineers, no overnight staff, and 70 mall sites with shared landlord fiber. Constraints are not negotiable unless the business changes them.
Two design principles follow. First, use an application-aware overlay that sends SaaS and guest traffic directly to the internet while steering POS and ERP to the data centers; this matches the traffic profile and reduces backhaul. Second, standardize with zero-touch provisioning and a single policy model because three engineers cannot hand-configure 320 stores.
Push back on dual internet at every store until site surveys prove two independent physical paths. At mall sites with shared landlord fiber, two carriers may share the same conduit, so the second link does not deliver the intended availability. Also challenge 99.95% POS uptime if LTE is the only backup and has a data cap; state the dependency explicitly.