What bandwidth and transfer each measure
The order form shows two words that do not measure the same thing. Bandwidth, written in Mbps or Gbps, answers how fast the port may go in one second. Transfer, written in GB or TB, answers how many bytes moved in an hour, a day, or a month.
Treat the port as a speed limit on a lane, and the monthly transfer as the distance driven. A limit that stays put still allows a larger distance when the lane stays busy. Stopping the meter on distance does not take the speed sign down.
On the invoice, unmetered means the monthly byte total is not a separate charge. It does not turn the port into an open rate. A 250 Mbps tier is still 250 Mbps in any one second. The 1 Gbps and 2 Gbps tiers work the same way: the cap changes, and so does the monthly add-on. None of the three tiers bills transfer again.
A speed-test page tries to fill the port for a few seconds. A result near the purchased tier shows that the path could reach that rate for that short run. A result far below it is a reason to check the test location and whether the run was download or upload, before treating the port as smaller than ordered. One test neither proves the month needs that rate nor replaces the counters on the machine.
How fast each port tier is
Convert with decimal units. 1 Mbps is one million bits per second. Divide by 8 to get megabytes per second. 250 ÷ 8 = 31.25, so a full 250 Mbps port is about 31.25 MB/s. Count 1 Gbps as 1000 Mbps, about 125 MB/s. 2 Gbps is about 250 MB/s.
A 1 GB file, counted as 1000 MB, takes about 32 seconds on a full 250 Mbps port. The same file takes about 8 seconds at a full 1 Gbps and about 4 seconds at a full 2 Gbps, if the whole path, the far end, the disk, and the application all keep up. When the disk or the far end is slower, the port is not the thing at the cap, and the file will not shrink to those seconds.
Estimate a full month with the same units. A port held at line rate for 30 days:
TB = Mbps × 86400 × 30 ÷ 8 ÷ 1000000
86400 is the number of seconds in a day. The product is bits; divide by 8 for bytes, then by 1,000,000 for TB. Here 1 TB = 1000 GB and 1 GB = 1000 MB, matching the MB/s figures above. Substitute 250, 1000, and 2000 and the totals are 81 TB, 324 TB, and 648 TB. Those totals are what a port produces only if it stays full for all 30 days. They show how much distance the unmetered line covers. Whether the business needs a larger port is a question for the busy-window samples in the next section.
| Port | Line rate | Full for 30 days | Monthly add-on |
|---|---|---|---|
| 250 Mbps | 31.25 MB/s | 81 TB | $0 |
| 1 Gbps | 125 MB/s | 324 TB | $245.0 |
| 2 Gbps | 250 MB/s | 648 TB | $742.7 |
The negotiated speed of the NIC is a different number. On Linux, /sys/class/net/<nic>/speed often reads 1000 or 10000, which is the link between the NIC and the switch, 1 Gbps or 10 Gbps. That file does not return the purchased port. A machine can have a 10 Gbps NIC and a 250 Mbps port. Checking ethtool or this speed file to see whether the order is 250 Mbps will not match, because the two limits are not the same point.
Step up when the busy window stays full
A larger port is justified when the business peak is waiting on the port. A site that usually moves a few megabits a second, then uploads at a steady 250 Mbps through a backup window, is waiting on outbound. Visitors downloading a package while the inbound side sits on the cap are waiting on inbound. Compare each direction with the port number on its own. Adding inbound and outbound, then comparing that sum with 250, stacks two directions into a number the port was not stated as.
Put the samples in the window that is actually busy, not in a quiet afternoon test. Backups often run at night, releases sit in a fixed hour, and visitor peaks follow the business time zone. Short samples in a row show how long each direction stays high. Touching the cap for a second or two and falling away is a burst. When the higher direction sits near the port number for the whole backup or visitor window, that window is being stretched by the port, and the tier is short.
Sixty percent is a reading line, not a line on the invoice. Sixty percent of 250 Mbps is about 150 Mbps, of 1 Gbps about 600 Mbps, and of 2 Gbps about 1200 Mbps. In the busy window, if the higher direction spends most samples under that line, the port still has room. If it often crosses the line and repeatedly sits on the port number, look at the next tier.
| What the busy window shows | What to do |
|---|---|
| The higher direction stays under 60 percent, with rare touches that fall away | Keep the current tier |
| Outbound or inbound sits on 250 Mbps through a backup, sync, or distribution window | Look at 1 Gbps |
| The port is already 1 Gbps and the same kind of window still sits on 1000 Mbps | Look at 2 Gbps |
| A speed-test page fills the port once, while business counters stay low at peak | Keep the current tier |
| The disk or the far end fills first, and port samples stay well under the tier | Work on the disk or the far end; leave the port |
Moving from 250 Mbps to 1 Gbps multiplies the line-rate byte speed by about four, because 1000 ÷ 250 = 4. Moving from 1 Gbps to 2 Gbps multiplies it by about two again. A file shrinks by that factor only when the disk and the far end can absorb it. When samples already sit on the cap and the disk still has spare throughput, the next tier shortens that window. When samples stay far under the tier and the disk is already full, a larger port leaves the file time where the disk put it.
Read the counters after the machine is delivered
Read the NIC that holds the first address in the console. List the addresses, then replace the interface name below with the one that holds that first line. eno1 is only a placeholder.
ip -br addr
A ten-second sample is enough to see whether this short stretch is sitting on the port. Subtract the starting receive and transmit byte counters from the ending ones, divide by 10 for bytes per second, then multiply by 8 and divide by 1,000,000 for Mbps. Inbound and outbound print on separate lines.
DEV=eno1
RX1=$(cat /sys/class/net/$DEV/statistics/rx_bytes)
TX1=$(cat /sys/class/net/$DEV/statistics/tx_bytes)
sleep 10
RX2=$(cat /sys/class/net/$DEV/statistics/rx_bytes)
TX2=$(cat /sys/class/net/$DEV/statistics/tx_bytes)
awk -v rx1="$RX1" -v rx2="$RX2" -v tx1="$TX1" -v tx2="$TX2" 'BEGIN {
printf "in %.1f Mbps\n", (rx2-rx1)/10*8/1000000
printf "out %.1f Mbps\n", (tx2-tx1)/10*8/1000000
}'
Repeat every minute or two inside the busy window and keep the set of readings for the higher direction. A single sample is weaker. Sample a night backup after it starts, and a daytime audience during its busiest hour. A sample near zero while the machine is idle only shows that nothing was moving. It does not show that the port is larger than the work.
To see the shape of a whole day, install vnstat and let it cover at least one full busy day, then read it by hour. The first hour after installation has no history, so it cannot describe yesterday.
vnstat -i eno1 -h
On Windows, take the same ten-second difference on the adapter that is up and whose address matches the first console line. Replace the name with that adapter from Get-NetAdapter.
$n = "Ethernet"
$a = Get-NetAdapterStatistics -Name $n
Start-Sleep -Seconds 10
$b = Get-NetAdapterStatistics -Name $n
"in {0:N1} Mbps" -f (($b.ReceivedBytes - $a.ReceivedBytes) * 8 / 10 / 1000000)
"out {0:N1} Mbps" -f (($b.SentBytes - $a.SentBytes) * 8 / 10 / 1000000)
Check in this order.
- 01 Confirm the order is 250 Mbps, 1 Gbps, or 2 Gbps. The NIC speed file is not part of this step.
- 02 Find the NIC or adapter that holds the first address in the console.
- 03 During backup, sync, or the visitor peak, take several ten-second samples and keep inbound separate from outbound.
- 04 If the higher direction stays under 60 percent of the tier, keep it. If the whole window sits on the port number, look at the next tier.
- 05 If the port samples stay low and the job is still slow, inspect the disk and the far end before raising the port.
How the port add-on enters the monthly total
The port tier is added to the machine’s monthly total together with memory, disks, addresses, and the operating system. Multiply that sum by the number of months in the term: 1 for monthly, 3 for quarterly, 6 for half a year, and 12 for a year. No term is discounted, and there is no setup fee. None of the three tiers bills transfer by the terabyte, so a step up adds the port’s monthly price, not a usage balance.
The 128GB machine is $454.8 per month. The default pair of 1TB NVMe disks is direct-attached, with no hardware RAID, so usable capacity is the sum of both, 2TB. Leaving the port at 250 Mbps adds $0. Moving to 1 Gbps adds $245.0 per month, and this machine’s monthly total becomes $699.8. The quarter is $2,099.4 and the year is $8,397.6. Moving to 2 Gbps adds $742.7 per month. With no extra disk, no extra address, and a free operating system, the monthly total is $1,197.5. The 256GB machine is $612.7 per month, with DDR4-3200 ECC RDIMM; the memory tier is chosen on the order. The processor is an AMD EPYC 7543P, 32 cores and 64 threads, 2.8 to 3.7 GHz. Each added 2TB NVMe disk is $84.7 per month, at most two, and each stays direct-attached. Each extra IPv4 is $1.6 per month. The machine is deployed in Tokyo, Japan. Linux is reached with SSH. Windows is reached with RDP.
When a disk, an address, or a paid operating system is added, fold those monthly prices into the figure that already includes the port, such as $699.8 or $1,197.5, and then multiply by the months. Multiplying only the port add-on, while the machine itself is still priced as one month, puts the term out of line.
On the configuration page the port is one of three choices. Use the busy-window samples to pick the tier, then use this section to multiply the add-on into the term. The monthly byte column is already covered by unmetered transfer. The choice itself only answers whether the port is fast enough at the peak.
Read the peak, then pick the port
$454.8 / month upward