In this guide:
- Measuring utilization honestly: peaks, percentiles, and busy hours.
- Trend-based forecasting, and the application-level drivers that bend trends.
- The cases where traffic shaping beats an upgrade.
Percentiles over averages
Averages flatter every link. The honest figures are the peak, the 95th percentile, and the busy-hour profile. The 95th percentile (discard the top 5 percent of samples, take the highest remaining value) is the industry's standard measure of sustained load, and the basis on which burstable bandwidth is commonly billed, which makes it the natural common language between your planning and your carrier's invoice.
Measure per link, per direction, at resolution fine enough to preserve the daily shape. A circuit whose 95th percentile sits at 80 percent of capacity with a rising trend is a planning item now, whatever its average claims.
Forecast from trend
With honest measurements retained over months, forecasting is a line through the data: at the current growth rate, this link's 95th percentile crosses the planning threshold in N months. Procurement lead time then converts N into a decision date, since a circuit that takes 90 days to upgrade must be decided 90 days before the crossing, at minimum.
Two practices keep the forecasts sound. Forecast per link rather than in aggregate, because growth concentrates unevenly and the aggregate hides the three circuits doing all the growing. And keep at least 6 to 12 months of history, so the trend line sees seasonality (quarter ends, seasonal business cycles) rather than mistaking a seasonal peak for a permanent slope.
Watch the drivers, because trends bend
Utilization trends are the output; application adoption is the input. New demand arrives as categories before it arrives as saturation: a video platform rollout, a cloud migration redirecting internal traffic through the interconnect, a backup redesign, an office expansion. Application-level and group-level traffic views catch these inflections while they are announcements rather than emergencies, which is the entire difference between planning and reacting.
This is also where planning meets the rest of the monitoring practice: the same conversation-level data that answers security and troubleshooting questions attributes growth to its causes, and growth with a named cause can be planned, shaped, or challenged.
When the upgrade is the wrong answer
Rising utilization has more than one remedy, and buying bandwidth for undesired traffic is the expensive way to avoid a conversation. Before any upgrade, attribute the growth: which applications, hosts, and destinations drive it. Recreational streaming saturating a branch uplink is a policy and QoS matter. A backup job overlapping business hours is a scheduling matter. Cloud sync storms are a configuration matter. Shaping, scheduling, and prioritization resolve a meaningful share of capacity pressure at configuration cost, and the traffic data that justifies an upgrade is the same data that identifies the cases where none is needed.
The honest planning posture holds both tools: upgrade where attributed business demand fills the link, shape where attributed non-business demand does, and let the attribution decide rather than the discomfort.
Capacity planning with ManageEngine NetFlow Analyzer
ManageEngine NetFlow Analyzer retains the per-link, per-application traffic history that capacity arithmetic runs on, and forecasts from it.
Feature highlights:
- Percentile and peak reporting: the honest figures, per interface and direction, over configurable periods.
- Trend and forecast views: growth trajectories from retained history, per link.
- Application-level attribution: name the drivers behind every growth curve before funding them.
- Scheduled capacity reports: monthly trend deliverables timed to budget cycles.