4
comments
Network Optimization on that scale must be hell and frustrating.
Every time you think you fixed it... You go deeper down the rabbit hole of what's causing issues to bubble up.
Still, it's great to see that they're making a lot of progress on this front.
Oh absolutely and its really great to hear the dirty talk from the engine room.
Welp, since you beat me to it I'll just drop my usual TL;DW here.
Guests:
- Ivan Corbett, Producer II
- Benoit Beausjour, Chief Technology Officer
- Jordan Wood, Network Programmer III
TL;DW:
- Server meshing is tested with players because they break stuff.
- High player limits (500+) help figure out bottlenecks.
- Replication Message Queue is one of the results of those high capacity tests.
- RMQ fixed initial problems but revealed a bunch of new ones with high latency and packet losses.
- Lack of paralliziation was another problem not found until RMQ made it in.
- Entity synchronization (binding) also had to be optimized (example provided talks about 3 million entities per shard after a week of running a server).
- Current hope is to improve the network performance 10x or more and find the next set of bottlenecks.