A Studio Session for Danko, a Slow-Internet Mystery, a New AI Council, and a Batch of OmniBridge Fixes

The slow-internet mystery

A speed test on our router kept showing about 7 Mbps down and 16 Mbps up on our own internet address space, even though the connection underneath is capable of far more — our provider for that space guarantees at least 200 Mbps. I spent a chunk of today chasing it down, and the fix sent that same speed test from 7 down / 16 up to 303 down / 42 up — comfortably above what we're owed, and our connection stayed up the whole time I made the change.

Getting there wasn't a straight line. I checked both ends of our setup first and found nothing broken. Asked the council of AI models I've started using for a second opinion, and most of them guessed our provider was throttling us on purpose — I almost opened a support ticket over it before testing proved that guess wrong. The real clue: traffic on low network port numbers moved at full speed, while traffic on high port numbers got cut down to a fraction of that. We'd hit this exact thing before with our VPN software, which is why those connections had already been moved to low ports — one of our other connections had never gotten the same fix, so it was still picking a random port each time and landing on one that got throttled. Gave it a fixed low port instead, same as the others. Worth admitting: I also made one wrong claim along the way, caught it myself, and corrected course before it went anywhere. Once the fix worked, I applied it to our other connections too, so nothing else can hit the same ceiling.

Also turned a new experimental option for our VPN software into a real installable package tonight instead of a hand-copied file, so testing it is less of a hassle. Still experimental, still off by default.

A batch of OmniBridge fixes

Also shipped three releases on OmniBridge today, chasing down bugs that actually affect how it reads during a live show:

• A receive-only connection was always showing 0 ms for round-trip time (how long it takes a signal to go out and come back), which looked like a perfect connection when really we just didn't have the number. Now it shows a dash instead of a wrong number.
• Changing a source's settings restarts its connection, and every so often the new one tried to grab the same network port before the old one let go of it, killing the source until a restart. Now it retries for a few seconds first.
• The dashboard was counting duplicate re-sent packets as dropped packets, so a healthy encoder that resends a lot of data looked like it was losing video. Split those apart so "Dropped" only means an actual loss.
• A connected source that wasn't routed anywhere yet showed as a black tile instead of a live picture, so an operator couldn't see what was coming in before routing it. Fixed.
• Added a read-only connection to Epiphan's encoders, so OmniBridge can show their own numbers (round-trip time, resent packets, delay) right next to what it's already measuring on the receiving end — handy for spotting whether a problem is on their side or ours.