I love the smell of

in the morning
Everything here is my opinion. I do not speak for your employer.
January 2013
February 2013

2013-01-19 »

Every now and then I remember my internship in fall 2000, at a startup where we were doing real-time audio and video over a multicast-like unicast protocol my friend had invented, which avoided actual multicast because actual multicast requires cooperation from all the intervening networks.  My job was to write the software and firmware, including some FPGA code, for the GPS PCI card that we designed to provide the time signal.  It turns out if you sync your media clock within 20 microseconds or so of GPS (this is safely within the available GPS precision), you can choose your packet boundaries based on the timestamp and guarantee that particular live digital audio samples will always be in the same spot in the same packet, even between independent/redundant packet sources ("nework transmitters") that don't talk to each other.  And you can reliably add sequence numbers to those packets that will be identical between transmitters.  From there, it's a "simple" matter to distribute the packets over redundant network links and at each router along the way, take the first packet you receive with a given sequence number, then retransmit to your three downstream peers.

The net result is that adding redundancy reduces average latency (since you take the first match, you don't wait for voting), it virtually eliminates any need for retransmission (since 99% reliability is 1% packet loss, and for three independent network paths times, that makes 0.0001% loss), and only costs 3x as much as a single feed.  When you're distributing live data to, say, 1000 people, 3 feeds is not expensive compared to 1000 feeds.

Sadly this company was a .com victim (their primary customers, including Intel's Media business, were shut down shortly after my internship), so this technology never left the beta stage.  They did okay though, moving into medical imaging distribution and storage, and eventually getting acquired by some famous storage company for lots of money.

12 years later, when I see ill-conceived proposals for how to distribute video over the internet, mostly they just make me sad.

I'm CEO at Tailscale, where we make network problems disappear.

Why would you follow me on twitter? Use RSS.

apenwarr on gmail.com