DOTS NetCode for MMORPG

What if you just not gonna use the rollback, no predicted ghosts. Will it still run some rollback related systems and generate rollback data for all objects?

I was planning to do exactly that, not use any prediction, my game does allow it. But now I see why they still might be generating rollback info. I hope this won't happen, or can be disabled.

 

Why not 1000 thin clients? Can't you run those like from a single laptop? Isn't that the same test for a server?

 

Sure.
I want it to be as real world as possible though so there is absolutely no doubt

 

We already have a bandwidth debugging tool in DOTS NetCode which runs on the browser but shows you a graph of upstream and downstream bandwidth for each network tick and ...

Regarding animations. You almost never send anything that you don't have to send which includes animation so you never will have animations sent over the network unless you have a very specific case which you need to synchronize a special event which is related to the gameplay of something specific to your game but i cannot think of any.

Most of the challenges a MMO will face will not be solved automatically by any tool and at best tools and libs can give you tools like different algorithms to choose from and tune their numbers for area of interest and bandwidth priority management.

 

Guys I've been reading this fascinating blog different networking ideas with regards to Fortnight / MMO. Given we are now at ECS 1.0 pre15 and we have a Netcode for Entities to support it, does this change anything? Are we at the stage where it could handle a 1K CCU on a single dedicated high end server streaming a world on a 10Gbps backbone like a 1km city scale? I'm just interested how close to this is achievable?

Some other things i'm interested in from a networking perspective would be

Oceans
Skies
Terrains
Mesh Rigged based Animations
GUI

Your thoughts much appreciated guys

 

Netcode for Entities, while it does add some cool new features, doesn't fully solve the concurrent users in one area problem. I'm sure it can handle a lot of players at once, maybe even a thousand because of the priority accumulator, but things reaaally start to bog down once players decide to throw a cool party in one area of the map. The "how many players does x support" is very difficult to answer because it depends on several factors like Total Players Online, Average Player Density, Update Frequency and Average Snapshot Size.
However, DOTS does improve the CPU usage by an insane amount so I suggest testing out the limits.

All of these should be simulated on the client and not networked. If you are doing voxel terrains then there are compression techniques for that but Netcode for Entities doesn't include those features, you have to do that yourself.

 

In my experience you can push vanilla NetCode to about 200 users. To start to get an idea of exactly how many you could get I recommend opening up NetDbg and playing around with your ghosts. We're absolutely network bound when talking about netcode in ECS (as opposed to CPU bound).

The first optimization you could make is to implement the under-documented "Relevancy" feature that is implemented here in the Asteroids sample. This makes it so that far away entities are not sent to a player that they are not relevant to. This is a necessary step to building a large world.

By default ghosts send their position and rotation constantly. We can adjust the rate with things like importance and even distance-based-importance but these are settings specific to your use case so hard to talk about in the abstract. However when networking a large number of entities we'd ideally only send the "inputs" and then calculate the same outcomes on the clients and server. We can sort of do this by using GhostVariants to not send the position/rotation data and manually sending it at an interval (or not at all) but I don't have the chops to tell you exactly how this would work with things like rollback and whether or not it would actually produce the same simulation on the client and the server after a long enough time.

No, the only true answer is a multi-server setup. The only general purpose, publicly available example of this that I've seen is SpatialOS. However it seems like SpatialOS kinda sucks. For that reason I've proposed a Unity-ECS-based SpatialOS-like here. It seems to me that something like this would be very possible to make for 1 person with the right background and inclination to make it. If it were made, it could be applied generally to make scaling realtime games with features like rollback, and I think that'd be pretty cool.

 

SpatialOS for Unity is completely abandoned and they massively overscoped their integration. Coherence rised up to take their place but doesn't seem to focus on the ECS workflow and the "advanced" concepts that Netcode provides.

I just want to chip in, as an ex SpatialOS poweruser and a creator of a similar package as Netcode that I have very similar thoughts. If we can glue together two servers, even in a crude way, it goes a long way.

 

I know this is an old thread but I think the discussion is important still.

What is the max client connection count?
- When I use Netcode for GameObjects?
- When I use Netcode for Entities?

I heard that Unity Netcode does not suit for MMO games.
Should I consider using other means to make a MMO game?

 

The short answer is, yes, you should use other means to make an MMO. The long answer is below.
Unity DOTS is very fast, but it's about efficient use of CPU cycles. Network performance is almost completely disconnected from CPU performance. With that in mind, almost any single-server setup (i.e., netcode for gameobjects, netcode for entities) is going to face the same limitations when it comes to CCUs. The exact number is almost impossible to say as it involves a lot of considerations about your games design, but almost any kind of interactive game will end up hitting the limit well before 300 CCU in "worst case" scenarios.

What do I mean by that? Well with either netcode you're sending updates about the state of relevant entities to every player every update interval. That update can only be so big, so what information goes in there matters. Entities with lots of ghost fields that change all the time are going to be more taxing to send. The more entities that are "relevant" the less information about each entity that will be able to make it into each update.

So we can get more CCUs by having fewer ghost fields and by having less relevant entities.

Again, no matter how much cutting you do you eventually hit an upper limit, which I estimate to be well below 300. If you want to break that you need a multi-server architecture. SpatialOS is an "out of the box" multiserver architecture, but I don't know much about it. Good luck!

 

okay I know what you mean.
Thank you very much for your clear explanation.
Is there any a method I can simulate my NGO game project is working with 256 players?
Can I test with Multiplayer Tools?

 

I am very happy to hear that.
Hope talking more about this.
I just hope more or less 256 CCU

 

There's a MMO game I know able to handle 15000 players at the same time at 64 core 128GB memory game server powered by .NET 6. Since dots netcode using dots tech, I guess it will able to handle even more players with the same game server spec?