Networking Physics / Lockstep VS State sync (Next Gen Networked Games Episode 2)

Рет қаралды 10,900

J. P. NextGenNetGames

Күн бұрын

Пікірлер: 18

@Katzberg-Damk Күн бұрын

As an game developer I think this video is 10/10. Thank you very much I m learning a lot

@pixboi Күн бұрын

There's one thing about lockstep => for the sim to work, everyone needs the same information=> this makes it hard for games, where you're not supposed to know some information => hacking is easy. With the normal snapshot interp, the server can decide what to reveal.

@pixboi Күн бұрын

Also, the determinism isnt "solved with a checksum", it requires often a bespoke fixed point math physics engine to work. And that also limits the fidelity/accuracy of the physics simulation.

@NextGenNetGames Күн бұрын

@@pixboi Yeah, a bit of a misquotation of the original thesis here, one of many in the series at this point :D, it was supposed to be: "Given the same inputs, the outputs are the same. We need a way to guarantee this is the case, this is solved with a checksum ", along with some more info on random numbers etc. As to determinism itself, honestly these days with ever more standardized platforms it has become a lot easier, we did tests on Bullet for Arma Reforger, and found that even complex simulations like a fully physically simulated vehicle were fully deterministic for multiple hours on average (after significant modifications of course), across platforms, on different builds (headless vs server vs client). You can check it out at: reforger.armaplatform.com/news/server-authoritative-vehicles Full lockstep determinism is another challenge all to itself, honestly more so in how the entire engine is structured than in physics, and it takes a special kind of team to make it happen.

@pixboi Күн бұрын

@@NextGenNetGames Im currently building a lockstep with physics, having my own c# dll that has fixed point math and unity to show stuff. it works well for basic usage, but ill have to see how far it goes

@Hellishcaos Жыл бұрын

Really good Video. Currently trying to learn about lockstep implementation so I can make my own RTS work properly. Doubt you'll see this but if you know any good places to learn about implementing and coding deterministic lockstep. Feel free to let me know.

@NextGenNetGames Жыл бұрын

I would recommend reading some of the Age of Empires articles I mention in the wiki, they are quite insightful. As far as implementation, I haven't looked at it personally, but I heard 0 A.D. is quite okay, its an open source online lockstep strategy game. I would also recommend "Detecting Synchronisation Problems in Networked Lockstep Games" by HAMPUS LILJEKVIST , I believe it is freely available online, it is a nice overview of the issues you will encounter. Personally I am still on the fence about truly recommending the technique, even for RTS. Beyond the technical concerns (even the greats struggle with it kzbin.info/www/bejne/gIipqmqvi7-Ui80 ) and hacking concerns (impossible to have true fog of war) it requires repeated reliable transmission a.k.a. it cannot be bandwidth bound, as far as I know there is no good solution to this, meaning that a client with a poor connection/bandwidth cannot play the game.

@kinggore5229 2 жыл бұрын

UE4 doesn't apply lockstep in anyway, dude.

@Trait74 2 жыл бұрын

Severely Underrated video and provides really valuable knowledge, THANK YOU!

@fever081 2 жыл бұрын

How UE Chaos can possibly implement all those types of networking if it simply implements the most naive physics replication without any prediction out of the box? There is no even quantization or other optimizations for net serialization. Network Prediction plugin is immature and forces to use fixed engine frame rate and also switches physics to run on the game thread. Async model is abandoned at the bleeding edge version at perforce. GeometryCollection for destruction also has a placeholder replication (no optimized net serialization) that is broken at 5.0.2. There are virtually none of those techniques implemented in base UE4/5, don't confuse people please.