Data centers and IP transit carriers often bill bandwidth by the "95th percentile", also called 95th percentile or burstable billing. It sounds complicated, but the rule is simple: measure your traffic rate regularly, throw away the busiest 5% of measurements, and bill the highest one that remains. This guide explains how the number is calculated, how it compares with flat-rate ports and monthly transfer plans, and how to read the traffic graph in your server panel.
In practice, short bursts are free. A backup that saturates the port for one hour every night adds up to about 30 hours a month, which is less than the 36 hours that get discarded, so it does not raise the bill. Traffic that stays high for more than about 36 hours in the month does.
| Usage pattern | Time at that rate | Effect on the 95th percentile |
|---|---|---|
| Normal traffic around 80 Mbps | Most of the month | Sets the baseline |
| Evening peak at 150 Mbps | 4 hours a day, about 120 hours | Longer than 36 hours, so the 95th percentile lands near 150 Mbps |
| Nightly backup at 900 Mbps | 1 hour a day, about 30 hours | Shorter than 36 hours, so it is discarded |
Providers measure in one of three ways, and your contract will say which:
| Model | What you pay for | Good for |
|---|---|---|
| Fixed port (dedicated / unmetered bandwidth) | A port speed such as 100 Mbps you can use 24/7 | Steady, predictable traffic; easy budgeting |
| Monthly transfer (for example 10 TB) | Total data in GB or TB | Low average usage with occasional spikes |
| 95th percentile (burstable) | A committed rate plus the 95th percentile above it | Large, variable traffic: CDN nodes, IDC resellers, streaming |
The difference between Mbps and data volume is explained in Mbps vs MB/s, bandwidth and monthly traffic.
Open your server in the client area and go to Server details → Traffic (see the dedicated server control panel guide):
The Live inbound / Live outbound rows in the Bandwidth & Traffic card show the newest sample, refreshed every 30 seconds.
If you have your own samples, for example exported from your monitoring as one Mbps value per line, the 95th percentile is a one-liner:
# samples.txt: one rate per line (Mbps), e.g. 5-minute averages for a month
sort -n samples.txt | awk '{a[NR]=$1} END {i=int(NR*0.95); if(i<1)i=1; print "95th percentile:", a[i], "Mbps of", NR, "samples"}'
To collect your own data on Linux, vnstat gives you a long-term history; Linux server monitoring shows how to install it.
rsync --bwlimit=20000 (KB/s).Not if it lasted less than about 36 hours in the month in total, because those samples fall into the discarded 5%. On a fixed-port plan a spike is simply limited by the port speed.
It is calculated only from the samples inside the selected range. A billing period is a full month, so compare the monthly value.
Some switch ports do not report traffic samples. Open a ticket and we will check the collection for your port.
Still stuck after following these steps? Open a support ticket and the IMIDC 24/7 technical team will help. Include the server IP and a screenshot of the traffic graph so we can pinpoint the issue faster.