All case studies
ISP / Carrier · 2025
Cutting peak-hour latency by 68% for a regional ISP
A growing ISP was losing customers to peak-hour congestion complaints despite having spare upstream capacity. A structured health audit found the bottleneck in the routing and edge layers, not the transit links.
Regional fibre ISP, ~14 000 subscribersGauteng, South Africa
- -68%
- Peak-hour latency
- 0.02%
- Peak packet loss (from 1.9%)
- -41%
- Congestion-related tickets
- 22%
- Transit cost avoided via caching
The challenge
- Evening latency spikes of 180–240 ms with intermittent packet loss on high-value business lines.
- No baseline telemetry: troubleshooting was reactive and driven by customer tickets.
- Asymmetric routing between two POPs caused traffic to hairpin through a single congested uplink.
- Single BNG handling all PPPoE sessions with no failover path.
What we did
- Deployed flow and SNMP telemetry across all POPs to establish a 14-day performance baseline.
- Rebuilt iBGP with route reflectors and corrected local-preference and MED policy to remove hairpinning.
- Introduced a second BNG with session redundancy and moved CGNAT onto dedicated hardware.
- Enabled FQ-CoDel style queue management at the edge and added on-net caching for the top content sources.
- Handed over Grafana dashboards, alert thresholds and a NOC escalation runbook.
Technology
- MikroTik CCR
- Juniper MX
- FRRouting
- LibreNMS
- Grafana
Facing something similar on your network? We start with a measured assessment, not a quote for hardware.
Book an assessment