Latency is how long data takes to cross the network, jitter is how much that delay varies from one packet to the next, and packet loss is the share of data that never arrives. A connection can have plenty of bandwidth and still fail on any of the three, and when it does, calls break up, video freezes and applications feel sluggish. Most problems start inside the office, on Wi-Fi or a full upload, rather than with the carrier, so measure before you blame the circuit.
A speed test tells you the size of the pipe. These three measurements tell you how smoothly it flows. They are the numbers that decide whether a voice call sounds like a conversation, and they are the numbers an SLA on a good business circuit actually promises.
Latency
What it is
Latency is the time it takes a packet to travel from one point to another, usually reported as round-trip time (RTT) in milliseconds: out and back. Ping, the most common test, measures round trips.
Where it comes from
- Distance. Light in fiber travels at roughly two thirds of its speed in a vacuum, about 5 microseconds per kilometre. That works out to roughly 1 ms of round-trip time for every 100 km of fiber route. Nothing can beat that.
- Route. Fiber does not run in straight lines. It follows roads, rail lines and existing conduit, and traffic may pass through a distant exchange point on the way.
- Queuing. When a link is busy, packets wait in line at routers. This is the part that changes minute to minute, and usually the part you can fix.
- Processing. Firewalls, VPNs and routers add a little time each, more when they are overloaded.
| Fiber route length | Round-trip time from distance alone |
|---|---|
| 10 km (across a city) | About 0.1 ms |
| 100 km | About 1 ms |
| 1,000 km | About 10 ms |
| 5,000 km (coast to coast, roughly) | About 50 ms |
These are physics floors, not measurements. Real round trips are higher, because routes are longer than the map distance and equipment adds time. The table is useful for one thing: if a server a few miles away shows 80 ms, distance is not the reason.
Jitter
Jitter is the variation in latency between packets. If one packet takes 20 ms and the next takes 60 ms, the stream is jittery even though the average looks fine. Voice and video suffer most, because they send a steady stream of small packets that need to be played back in order and on time. Phones and meeting apps hold a small buffer to smooth out variation; when jitter exceeds the buffer, packets arrive too late to use and are thrown away, which sounds exactly like packet loss.
Common causes are congestion that comes and goes, Wi-Fi retransmissions, and a busy upload where voice packets queue behind large transfers.
Packet loss
Packet loss is the percentage of packets that never arrive. For downloads and web browsing, lost packets are resent, so loss shows up as slowness. For live voice and video, there is no time to resend, so loss shows up as gaps, clipped words and frozen frames. Even small, steady amounts of loss are noticeable on calls.
Loss comes from overloaded links dropping packets when queues overflow, from Wi-Fi interference, from faulty cables, optics or ports, and occasionally from problems in a carrier's network.
What voice and video can tolerate
These are widely used rules of thumb, not guarantees. The ITU-T G.114 recommendation treats one-way delay up to about 150 ms as acceptable for most interactive use, and the other figures are common planning targets. Individual platforms publish their own requirements.
| Measure | Comfortable for voice and video | Where problems usually start |
|---|---|---|
| Latency | Under about 150 ms one way (roughly 300 ms round trip), and much lower is better | Conversations start to overlap as delay rises past that |
| Jitter | Under about 30 ms | Above the device's jitter buffer, audio breaks up |
| Packet loss | Under about 1% | Steady loss above that is clearly audible on calls |
On a good dedicated fiber circuit to a nearby server, you should see figures far better than these targets. Our guide to what a 99.99% SLA buys covers how carriers commit to latency and packet delivery in writing on dedicated internet access.
Why more bandwidth does not always help
A faster circuit fixes latency only when the delay comes from a full link. It does nothing about distance and nothing about Wi-Fi interference: a 1 Gbps circuit to a server on the other coast faces the same physics as a 100 Mbps one. Where the delay comes from queuing on a busy upload, more capacity helps directly, and so does managing the queue so that voice goes first. This is also where SD-WAN earns its place. With two circuits, it can send calls down whichever one is showing the lowest latency and loss at that moment, rather than trusting either one blindly.
Where the problems usually are
The carrier circuit is often not the culprit. Work from the inside out:
- Wi-Fi. Interference, crowded channels and weak signal cause jitter and loss. Desk phones and conference rooms should be wired where possible.
- A full upload. When uploads saturate the link, every outbound packet queues, and latency climbs for everything. See symmetrical vs asymmetrical internet.
- Your own equipment. An overloaded firewall, a failing switch port, a damaged patch cable or an old router can all drop packets.
- The access circuit. Shared services can congest at busy times; a fault on a dedicated circuit usually shows up as loss at all hours.
- Beyond the carrier. Traffic may be routed a long way round to reach a particular service, or the service itself may be struggling.
How to measure it
- Test wired first. Plug a laptop directly into the switch or firewall to take Wi-Fi out of the picture.
- Ping over time, not once. A continuous ping for several minutes to a reliable destination shows latency, variation and loss. A single test shows almost nothing.
- Test under load. Run the same ping while someone starts a large upload. A big jump in latency points to queuing on your upload.
- Trace the path. Traceroute and tools that combine it with continuous ping show where along the path delay or loss appears. Many routers deprioritise answering these tests, so loss that appears at one hop and not at the hops after it is usually not real loss.
- Watch continuously. Intermittent problems need monitoring over days. SD-WAN devices measure every path all the time, and a managed network service can alert on changes before users notice.
A worked example
Hypothetical office and measurements, invented to show the diagnostic process.
A 15-person office complains of choppy calls every afternoon. The circuit is a 300 Mbps service, and speed tests look fine. The team runs five-minute tests to a nearby server.
| Test | Round-trip latency | Jitter | Packet loss |
|---|---|---|---|
| Laptop on Wi-Fi, quiet morning | 18 ms | 12 ms | 0.8% |
| Laptop wired, quiet morning | 6 ms | 1 ms | 0% |
| Laptop wired, during the afternoon cloud backup | 190 ms | 45 ms | 0.5% |
Two problems appear. Wi-Fi adds jitter and a little loss on its own, so the desk phones on Wi-Fi are moved to wired ports. The bigger issue is the backup: it fills the upload, latency jumps and jitter blows past the phones' buffer. Moving the backup to overnight and turning on traffic shaping that gives voice priority fixes the afternoon calls. If the backup had to run during the day, the lasting fix would be more upload. At no point was the carrier circuit faulty.
Troubleshooting checklist
- Does the problem happen wired, or only on Wi-Fi?
- Does it happen at a particular time of day, or when a particular job runs?
- Does latency jump when the upload is busy?
- Is it one application or everything? One service suggests the service or its route.
- Do your firewall or SD-WAN graphs show loss on the circuit itself?
- Have you logged times and test results to give the carrier if it is theirs?
Phones are usually the first thing to show these problems. If calls are the main complaint, our business voice page covers what a clean call path needs.
Questions to ask your carrier
- Does the SLA commit to latency, jitter and packet delivery, or only uptime?
- Where are those figures measured, and are they monthly averages?
- Can I see performance data for my own circuit?
- Is my access circuit shared at any point, and how is it managed at busy times?
- If I report loss or high latency with test results, what is the troubleshooting process and how quickly does it start?
A connection judged on the right numbers
When we compare quotes, we look past the headline speed to the SLA terms on latency and packet delivery, because those are what your calls depend on. Tier 1 carriers bid on your address, the carrier pays us so nothing is marked up, and the quote is free. If your problem turns out to be inside the office, we will tell you that too. Send us your address, call 478-758-8091 or text (347) 870-0965, and you will usually hear back the same day.
