Network case study · September 2026
Rebuilding the Bangladesh Office Network
Wired connections were stuck at about 92 Mbps on every floor and Wi-Fi was weak. The first job was finding out why; the second was replacing the firewall, switches, access points and the wrong cable, without costing staff any working time.
Old equipment and a speed limit nobody had explained.
The office ran a SonicWall firewall whose security subscriptions had lapsed years earlier, Cisco switching and Cisco access points on a single flat network. Staff, guests and cameras all shared it.
- Remove end-of-life equipment that no longer received security updates
- Lift wired speeds stuck at about 92 Mbps on every floor
- Fix weak Wi-Fi, especially on phones
- Separate staff, guest and camera traffic
- Replace non-standard cabling
The bottleneck was physical before it was logical.
A limit of about 92 Mbps on every floor is a clue in itself: it is what a 100 Mbps link delivers in practice. Tracing it found two causes.
Every wired connection was capped by the switch, whatever the cable or the internet link could do.
Four FortiSwitch 248E-FPOE switches, managed from the firewall.
Some runs used copper-clad aluminium cable made for CCTV, which does not meet Cat5e or Cat6 and often cannot reach gigabit.
About 70 m of that cable and the non-standard patch leads were replaced with Cat6.
A new firewall alone would not have fixed either problem.
Moving the office onto the regional design.
Before
After
Three internet links were combined under SD-WAN with health checks, and guests and cameras were blocked from reaching staff devices.
Most of the work happened before the cutover.
Every switch port planned in advance, so the cutover was a matter of following the map.
Written configuration plan and step-by-step migration plan.
The firewall was configured before it reached the office, with a written list of what remained on site.
Old configurations were exported first, and the old devices stayed on site as a fallback until testing passed.
Thursday evening to Sunday.
| Day | Work |
|---|---|
| Thu 24 Sep, from 17:45 | Old equipment removed; firewall and switches installed; cabling started |
| Fri 25 Sep (weekend) | CCTV-grade and non-standard runs replaced; switch, network and firewall configuration |
| Sat 26 Sep (weekend) | Access points installed; server room tidied; first wired and Wi-Fi tests |
| Sun 27 Sep (working day) | Validation with all staff in; Wi-Fi tuning; security policy checks |
Once testing passed, the old devices were wiped for disposal.
Clients were being pushed onto the busiest band.
After cutover, Wi-Fi was better but not as good as it should have been. Four access points had manual settings that ran the 2.4 GHz radio at full power, so devices clung to a band that was up to 73% busy while 5 GHz sat below 10%.
Overrides on four access points held clients on the congested 2.4 GHz band.
Removed the overrides and applied one profile with automatic power and channel selection; 2.4 GHz limited to channels 1, 6 and 11.
Filtering was also checked against the Nepal office, so both offices block the same things.
Measured before and after.
| Test | Before | After |
|---|---|---|
| Wired download (median) | 92 Mbps (4 tests) | 723 Mbps (15 tests, range 420–790) |
| Laptop Wi-Fi download (median) | 79 Mbps | 158 Mbps (18 tests) |
| Phone Wi-Fi download | 4–19 Mbps | 132 Mbps (1 test) |
What this migration taught.
The speed problem was cabling and switch ports; equipment alone would not have fixed it.
Arriving with a configured firewall and a port map turned the cutover into execution, not design.
Old equipment stayed on site, configured, until the new network passed testing.
The Wi-Fi problem came from individual settings; one shared profile fixed it.