Peak concurrency, not the sum of everything you own
The wrong way to size a connection is to add up every device in the house. Most of them are idle most of the time, and a household with forty connected things may have four that are actually moving data on a given evening. The right question is how much is happening at the same moment during the worst realistic hour, which is usually a weekday evening and occasionally a wet Sunday.
That is why every field on this page asks for a count at once rather than a total. Three televisions in the house and two of them ever on together means two, not three. The exception is anything that runs unattended: backup software, photo sync and cameras do not wait for a convenient moment, and they are running during the worst hour by definition unless you have scheduled them not to be.
Upload is the one that runs out
Most residential connections are strongly asymmetric, meaning the upstream capacity is a fraction of the downstream. How large a fraction depends on the technology, and that is not something to state as a fact here because it varies by connection type, by provider and by market. What is general is the consequence: households size their plan on the download number, because that is the number on the advertisement, and then run out of upload.
The activities that consume upstream are not the ones people think about. Video calls are symmetric, so a call costs upstream at the same rate as downstream. Backup and photo sync will use every bit of upstream capacity you leave available unless configured otherwise, and they run at night, and they run in the evening too. Cameras send continuously. Meanwhile the whole household is downloading, and downloads need a return path for acknowledgements, so a saturated upstream slows the downstream as well. That is the mechanism behind the most common complaint in this whole subject: everything is slow, the speed test says the plan is fine, and nobody can find the cause.
Where the biggest single number usually is
Naming the dominant activity is more useful than the total, because it tells you whether a bigger plan is the answer or whether one setting is. In most households the largest single line is high resolution video, and it is largest by a wide margin over everything else combined.
| Activity | Bitrate character | What actually constrains it |
|---|---|---|
| High resolution video | Large and sustained | Downstream capacity |
| Ordinary video | Moderate and sustained | Downstream capacity |
| Video calls | Moderate, symmetric | Upstream, and jitter more than either |
| Gaming | Very small, constant | Latency and packet loss, not bandwidth |
| Game and system updates | Enormous, bursty | Downstream, and everything else while it runs |
| Backup and sync | Whatever it is allowed | Upstream, unless rate limited |
Two rows of that table are worth reading twice. Gaming uses almost no bandwidth during play, which surprises people who have been sold a plan on the strength of it; what gaming needs is consistent low latency, and a fat connection carrying a large upload has poor latency regardless of its size. And updates are the largest single transfer that happens in most homes, which is why an evening can be fine for a month and then terrible for an hour with nothing having changed.
Bitrate on the application is not bits on the wire
Every figure a service quotes for its own stream is the payload. What crosses the connection also carries headers at several layers, acknowledgements travelling the other way, and retransmissions of anything lost. Twenty percent is a reasonable planning allowance for the difference and the field is editable because the real number depends on conditions. A link that is losing packets, which includes almost any congested wireless link, retransmits more and the overhead climbs.
The same distinction applies at the other end of the calculation. The speed on the plan is a headline and the speed you get is a measurement, so the usable share field exists to let you put the measured figure in. Measure it wired rather than wireless, because a wireless measurement tests the wireless as well and confounds the two problems. If you are not sure which of them you have, the Wi-Fi improvement guide opens with the test that separates them, and it takes about ninety seconds.
What to do with a total that does not fit
A shortfall on the download side is usually solved with a larger plan or a lower video quality setting, and the second one is free. A shortfall on the upload side is a different conversation, because on some connection technologies more upstream is not purchasable at any price, and the practical answers are rate limiting the backup software, scheduling it for hours when nobody is on a call, and turning off the sync client that is uploading a video library nobody asked it to.
There is also a case where the total fits comfortably and the house is still slow, and it is worth naming because this page cannot detect it. If the demand is well inside the plan and things are still bad, the bottleneck is somewhere between the connection and the device: the wireless, a congested channel, an old switch, or one device retransmitting. Coverage is worked out in the access point coverage calculator and channel contention in the channel width planner. If cameras are a large part of the load, the camera bandwidth calculator breaks that out properly, and anyone sending video out of the house should size that side with the streaming bitrate calculator instead of guessing.
Questions people ask
How much internet speed does a family actually need?
Less download and more upload than most plans are sold on. The download figure that matters is the concurrent peak, and for most households that is dominated by however many high resolution video streams run at once, with everything else adding a modest tail. The upload figure that matters is video calls plus whatever backup and camera traffic is running unattended, and it is far more often the binding constraint. Rather than trusting anyone number, put your own worst evening into the fields above and look at which of the two lines goes red first. The answer differs enormously between a household that streams and one that video calls all day.
Why is everything slow when nothing is downloading?
The usual culprit is an upload, which is invisible because nobody is watching anything upload. Backup software, photo sync, a cloud drive catching up on a large folder, or a camera uploading clips will all saturate a modest upstream and stay there for hours. When the upstream is full, the acknowledgement packets that downloads rely on queue behind the upload traffic, so downloads slow down too and the cause looks unrelated to the effect. Check what is uploading before buying anything: most sync and backup applications have a bandwidth limit setting, and setting it to a fraction of your upstream fixes the whole symptom.
Does a faster plan improve video call quality?
Only if bandwidth was the problem, and it usually is not. Video calls need a few megabits each way and are far more sensitive to consistency than to capacity. Jitter, packet loss and latency spikes are what make a call break up, and all three get worse when the connection is congested by something large, which is why a call degrades the moment somebody starts an upload rather than when the plan is small. A congested wireless link produces the same symptoms for entirely local reasons. Fixing the congestion generally beats buying capacity, and costs nothing.
Should I count every smart device in the house?
Not by bandwidth, because most of them use a negligible amount. Sensors, switches, thermostats and similar devices send tiny amounts of data occasionally, and a hundred of them together are a rounding error next to one video stream. What they can affect is the wireless, since every device consumes some airtime just by being associated and by talking to the access point, and a large number of chatty devices on the low band is a real capacity problem even though it is not a bandwidth one. That is an airtime question rather than a plan size question and belongs with the channel and coverage planning.
What percentage of my plan do I actually get?
Measure it rather than accept a figure, and measure it at the time of day the problem happens, because congestion upstream of your house varies through the day. Test from a device connected by cable, since a wireless test measures the wireless as well and will understate the connection. Run it more than once. What is worth avoiding is treating any advertised or nominal figure as fact, including the throughput a wireless standard is said to reach, which in the real world is a fraction of the headline for reasons that have nothing to do with your provider.