The 24-Hour ARP Window: Why One Manufacturer Moved 1,200 Devices Off Flat Networks
Archi3 min read
Contents
The decision on the table#
At a manufacturer with four plants and a head office, the network lead had a straightforward brief: cut telecom spend by 30 percent and make the network easier to manage. The incumbent carrier had quoted a three-year MPLS renewal at a number the CFO had circled in red. A systems integrator proposed SD-WAN over two commodity internet links, with a promise of 40 percent savings and central policy control.
What followed was a six-month detour through three myths that shape most enterprise network decisions.
Myth: SD-WAN replaces the wide area network#
Fact: SD-WAN replaces the transport underlay and the routing policy on top of it. The company discovered this when the pilot site, a 300-person plant, kept dropping its warehouse management system during shift changes. The integrator blamed the ISP. The ISP blamed the firewall. The actual cause was a flat Layer 2 domain that stretched across 1,200 devices, from programmable logic controllers to VoIP handsets. Broadcast traffic from a single misconfigured printer was flooding the segment.
SD-WAN appliances cannot fix that. The company had to segment the plant into operational technology and business VLANs before any transport change could deliver value. That work took eight weeks and cost more than the first year of SD-WAN licensing.
Myth: MPLS is inherently more reliable than internet#
Fact: MPLS offers committed rates and predictable latency, but the reliability gap depends on where the traffic goes. The manufacturer's plants ran two production applications, a cloud ERP and a local MES. The ERP traffic already went over the public internet through a regional gateway. Only the MES traffic stayed on MPLS, and it never left the metro area.
After measuring 90 days of latency, the team found the MPLS circuit carried 18 percent of total traffic and accounted for 61 percent of the telecom bill. The internet links at the same sites had 99.94 percent uptime against a contracted 99.99 percent on MPLS, a difference the plant manager called "invisible until the invoice arrives."
Myth: Segmentation is a security project#
Fact: At this company, segmentation was a networking project first. The OT team had resisted VLAN changes for years because a previous attempt had broken a legacy serial gateway. The compromise was a 24-hour maintenance window per line, staggered across weekends, with a rollback plan that kept the old flat configuration on standby.
The first cutover took 11 hours. The second took four. By the fourth line, the team had a repeatable checklist: map every device by MAC address, confirm the PLC firmware version, test the safety interlock path, then move. The company eventually moved 1,200 devices across seven VLANs without a single production stoppage.
What the company chose#
The final architecture was hybrid, not pure. Two plants with heavy MES reliance kept an MPLS circuit for one more year, at half the original bandwidth, while SD-WAN handled ERP and general traffic. The other two plants moved fully to dual internet with 5G failover. The telecom bill fell 34 percent in year one, and the network team gained a dashboard that showed per-application latency instead of per-circuit status.
Three costs did not appear in the integrator's proposal: , Segmentation labor: roughly 640 engineer hours across four sites., Spare hardware: two standby SD-WAN appliances for rapid swap, about $18,000., Training: a two-day workshop for plant technicians who now troubleshoot VLAN tags, not just cable faults.
Myth: The carrier is the bottleneck#
Fact: The carrier was the easiest vendor to manage. The hard part was internal ownership. For years, the plant managers controlled their own switches because IT could not guarantee response times during a production run. The network team had to negotiate a service-level agreement with the plants before it could standardize configurations. That agreement, not the SD-WAN contract, made the project work.
The takeaway for other buyers#
A transport refresh is a good moment to audit what actually runs on the network. The manufacturer found 214 devices it had not inventoried, including 40 that had been decommissioned but still answered on the flat segment. The 30 percent savings target was met, but only because the team treated segmentation as part of the project, not an afterthought.
Companies weighing SD-WAN against MPLS should ask three questions before signing: how much traffic leaves the metro, how many devices share a broadcast domain, and who owns the switch configurations at each site. The answers will shape the contract more than any carrier's marketing deck.
More articles