DEV Community

ADAPT INDIA NETWORKS
ADAPT INDIA NETWORKS

Posted on

Your Speed Test Is Lying: How to Measure Jitter, Packet Loss and Bufferbloat on Office Internet

Every time someone complains that video calls are freezing, someone else runs a speed test, sees a big number, and says the internet is fine.

Speed tests measure one thing: how much data you can push through in a short burst. Calls, VoIP, remote desktop and most real work care about something else entirely, which is how consistent that connection is. A line can show 200 Mbps and still make every Teams call sound like a robot.

I work with a telecom partner in India and spend a lot of time explaining this to office IT folks. Here's how to actually measure what matters, with tools you probably already have.

The four numbers that matter

Latency: how long a packet takes to get there and back. Lower is better.

Jitter: how much that latency varies from one packet to the next. Calls hate jitter more than they hate high latency.

Packet loss: packets that never arrive. Even 1 to 2% is enough to make calls choppy.

Bufferbloat: how much latency jumps when the line is busy. This is the sneaky one, and it's why calls break the moment someone starts a big upload.

Rough rules of thumb for voice and video: latency under about 100 ms to the service you use, jitter under about 30 ms, packet loss under 1%. Not hard limits, but good enough to know if you have a problem.

Test 1: a long ping, not a short one

A 4-packet ping tells you nothing. Run at least 100.

Linux / macOS:

ping -c 100 1.1.1.1
Enter fullscreen mode Exit fullscreen mode

Windows:

ping -n 100 1.1.1.1
Enter fullscreen mode Exit fullscreen mode

Look at the summary at the end. Packet loss is the percentage. On Linux/macOS, the mdev (or stddev) value is a decent stand-in for jitter. On Windows, look at how far apart the minimum and maximum are. If min is 8 ms and max is 300 ms, you have a jitter problem even if the average looks fine.

Run it once when the office is empty and once at the busiest hour. The difference is the story.

Test 2: find where the problem is with mtr

Ping tells you there's loss. mtr tells you where.

mtr -rwzbc 100 1.1.1.1
Enter fullscreen mode Exit fullscreen mode

On Windows, the built-in equivalent is:

pathping 1.1.1.1
Enter fullscreen mode Exit fullscreen mode

You get a list of every hop between you and the target, with loss and latency at each one.

How to read it without fooling yourself:

  • Loss at a middle hop that doesn't continue to the final hop is usually fine. Many routers deprioritise replies to these probes.
  • Loss that starts at one hop and continues all the way to the end is real. Note where it starts.
  • If it starts at hop 1 or 2, it's probably your own router, firewall or Wi-Fi.
  • If it starts a few hops in, inside your ISP's network, that's the screenshot you send to their support team.

Test 3: run it over a cable first

Before blaming the ISP, plug a laptop directly into the router or firewall with a cable and run the same tests. If the cable is clean and Wi-Fi isn't, your internet is fine and your Wi-Fi needs fixing. This one step settles about half of all "internet is slow" arguments.

Test 4: bufferbloat

Start a long ping in one terminal:

ping 1.1.1.1
Enter fullscreen mode Exit fullscreen mode

Then, in another window, start a big upload (a large file to Google Drive, for example) and watch the ping times.

If latency goes from 10 ms to 400 ms while the upload runs, that's bufferbloat. It explains the classic complaint: "calls are fine until someone sends a big file". There are also browser-based bufferbloat tests that show latency under load if you want a quick picture.

The fix is usually SQM or QoS on your firewall (shaping traffic slightly below line speed so queues don't build up), or simply more upload bandwidth.

Test 5: real throughput with iperf3

Online speed tests hit a nearby server, often inside your ISP's network, so they can look better than real life. If you have a server or VM outside your office (a cloud VM works), install iperf3 on both ends.

On the server:

iperf3 -s
Enter fullscreen mode Exit fullscreen mode

From the office:

iperf3 -c <server-ip> -t 30
iperf3 -c <server-ip> -t 30 -R
Enter fullscreen mode Exit fullscreen mode

The -R flag tests the reverse direction, so you see both upload and download. On a shared broadband line, run it at different times of day. If the numbers swing a lot, you're seeing contention, other users in your area sharing the same capacity.

What the results usually point to

  • Clean on cable, bad on Wi-Fi: fix the Wi-Fi. Access point placement, channels, too many users per access point.
  • Bufferbloat during uploads: QoS/SQM on the firewall, or more upload bandwidth.
  • Loss or jitter starting inside the ISP's network: take your mtr output to the ISP.
  • Fine at 9 am, bad at 4 pm, every day: contention on a shared connection. This is where a dedicated internet leased line helps, since the bandwidth isn't shared and upload matches download. We explain the difference in more detail in our leased line vs broadband guide.
  • Multiple sites with uneven links: SD-WAN measures latency, jitter and loss on every link continuously and moves each app to the healthiest one, which is these tests done automatically, every second.

Save your results

Whatever you find, save the output with the date and time. When you call the ISP with "100-packet ping, 6% loss starting at hop 4, every day between 3 and 6 pm, mtr attached", you get a very different response from "sir, the internet is slow".

Got a weird mtr output you can't read? Paste it in the comments.

I work with Adapt India Networks, a Tata Tele Business Services and ACT Fibernet partner in New Delhi.

Top comments (0)