Screen Sharing at 3 Mbps: A Field Test Across Four Countries

Four reviewers on connections between 3 and 6 Mbps shared their screens for a week, and the median frame rate during a design review was 4 frames per second. The number is low enough to explain why those calls ended with a screenshot instead of a decision.

The test covered a design review, a spreadsheet walkthrough, a code reading session and a document edit, each run twice by each participant. Nothing was installed. The same browser-based tool was used throughout.

What the connections actually were

Two participants worked from home on 5 Mbps and 6 Mbps lines, one from a co-working space on a shared 3 Mbps link, and one from a mobile hotspot that measured 3.4 Mbps down and 0.9 Mbps up. The upstream figure turned out to matter more than the downstream one.

Screen share is mostly an upload, which is the direction most home connections optimise least.

The four sessions

Design review produced the worst result. At 4 frames per second a cursor moves in visible jumps, so pointing at a detail is impossible and people describe locations in words instead. The spreadsheet walkthrough managed 9 frames per second and worked acceptably, because scrolling is a coarser motion.

Code reading at 12 frames per second was usable. Document editing was the surprise: 15 frames per second, the best of the four, because the visual change per second is small.

Where the time went

Each participant logged the number of screenshots taken during a session. The design reviews averaged 11 screenshots, the spreadsheet sessions 4, the code sessions 2 and the document edits 1. Screenshots are the tax a slow share imposes: they replace pointing.

Every screenshot also produced a message to explain what the image was supposed to show, which is where the real cost sat.

What helped

Three fixes worked on every connection. Sharing a single window instead of the whole screen cut bandwidth by roughly a third. Turning off the camera during the share recovered enough upstream to raise the frame rate by two to four frames per second. Lowering the share resolution to 720p made text blurrier but smoother, and people preferred smooth.

The hotspot user benefited most from the camera fix, gaining six frames per second.

What did not help

Asking people to close other applications changed nothing measurable, because the bottleneck was the link rather than the machine. Turning off the video of other participants helped slightly on the shared 3 Mbps link and not at all elsewhere.

The co-working space could not be fixed from inside the call, which is a structural problem rather than a configuration one.

Six weeks later

The team adopted two habits: window sharing by default, and camera off whenever a share matters. Screenshots per design review fell from 11 to 5. Median frame rate in design reviews rose to 9 frames per second.

The 720p trade-off in practice

Lowering share resolution helped the frame rate and hurt legibility in a specific way. At 720p, body text in a document is readable and a spreadsheet cell reference is not. Two participants reported reading numbers aloud instead of pointing at them, which worked but turned a visual check into a verbal one and doubled the length of the call.

The team settled on 1080p for spreadsheet work and 720p for design reviews, which is the opposite of what the frame rates suggest. The deciding factor was not smoothness but whether a number had to be read.

What a screenshot actually costs

Eleven screenshots per design review sounds trivial until the follow-up messages are counted. Each screenshot produced an average of 1.8 replies, and each reply kept the thread alive into the next working day. The review that was supposed to close in forty minutes was still receiving messages eighteen hours later.

Reducing screenshots from 11 to 5 shortened the average review thread by roughly a third. The frame rate was never the goal. The thread was.

Four reviewers on 3 to 6 Mbps once saw 4 frames per second and 11 screenshots. The same four now see 9 frames and 5 screenshots, on the same connections, with two settings changed.