Comment by jeroenhd
1 month ago
I kinda wonder what the packet loss story of UDP is like for simple protocols like these in the modern day. Clearly, it's good enough that this application runs without issue.
> Instead of bolting TLS onto every application protocol (HTTP, MQTT, CoAP, etc.), what if we secure the network layer once and then use simple protocols on top?
IPSec, coming back with a vengeance!
This also makes me wonder what takes less space in ROM, a basic WireGuard implementation or a basic, stripped-down IPSec implementation (with only the ciphers and configuration necessary for the server compiled in). WireGuard has the advantage of being designed for simplicity, but IPSec has its 90s every-cycle-counts legacy that a lot of modern software has ignored since.
> packet-loss
For one it can adequately show if your WiFi is performing well...
In a home network setting I've got a UDP display sink (64x64px, RGB) with a custom protocol. It works fine-- no hard number as the performance varies based on other factors, including congestion. I've been able to push it to >90fps, but around 42fps there is virtually no flicker.
For context in my protocol I use all 512bytes, with 3 first being the "line number and such" data and the rest containing ~128 pixels.
I monitor packet loss on both sides all day every day. It’s still a thing, but different than I’d imagined:
- Episodic most often. Something transient causes high loss for a short time. This happens locally and “in the cloud”. - Persistent due to a back connection or very high network load. I only really ever see this locally.
But I can go days at a time and not lose even one of millions of probes and responses.
When connections are good, they’re excellent these days. When they’re bad, well that doesn’t seem to have changed.
WireGuard is far smaller in my experience. I don’t set aside the possibility that someone more clever than me can get IPsec condensed down to something tiny but I never could — it’s just too much in one bag, so to speak.