Star Citizen, Roberts Space Industries - Chris Roberts' life support and retirement fund [2012-]

[CSI PC]

The physical network itself is hardly relevant, the necessary bandwidth is comparatively tiny and the latency you have to deal with is no different for 10 players as a 1000. It's a database and physics problem, how to do physics updates for 1000s of entities and how to generate+compress thousands of worldviews in a single tick. Also the client needs to be able to do good culling and LOD management of course.

Latency still relates to physical networks, not all global networks are equal and especially home users.
Then there is the behaviour of the content switching-load balancing/backbone setup/redundancy/etc.
But my point went well beyond that and physical network, and from experience in this subject involves those I mentioned; if looking to do an implementation and solution that is massive/global and critically efficient.

This game will still have time/session sensitivity that is seen in more general MMO arcade games, along with as you say database-server/client related application challenges.
Like I said this involves a lot of the OSI Layer to do well, which includes session/application.

I think we are mostly in agreement just approaching this at different ends of the OSI scale (engineering vs programmer perspective).
From my experience all of the layers matter, and had to help a few times in resolving time sensitive issues because they did not understand the protocol-session behaviour well enough with their solution and also implication of global networking - just a basic example.

 

[cheapchips]

SpatialOS, used in World's Adrift, look like just the type of third party network solution CIG could look at. Shame they locked themselves out of that by using Lumberyard.

 

[eastmen]

i know back when tI was still following the game closely they were using google servers. I wonder if it would be smart of them to go to microsoft. They have a massive cloud division which is easily top 3 (it might be ahead of amazon now) and they have alot of gaming knowledge. Heck MS has bailed Roberts out before. They could even bring in more money that way by putting SQ42 on the next xbox as a launch title ... at this point it wont be far out of line from a release at that time anyway

 

[cheapchips]

They can't use anything but AWS now they're on Lumberyard. It's the main condition that comes with using the engine.

 

[MfA]

Running massively multiplayer zones on AWS will make things far harder than it has to be ... I don't see them pulling it off this decade.

 

[CSI PC]

They will have to go similar path to Elite Dangerous IMO that combines AWS with good peer-to-peer.

 

[Jupiter]

Last years December Star Citizen backers asked for a net code special and they will get one on January 19th. 12pm PST.

https://www.robertsspaceindustries.com/spectrum/community/SC/forum/50259/thread/net-code-special

 

[CSI PC]

Did Frontier and Elite Dangerous start their development with AWS and so had plenty of time to build their solution around it?

 

[snarfbot]

man imagine if they used the kickstarter money to buy bitcoins instead and just sat on them till they were worth 19000 dollars and then sold them, christ that wouldve been a lot of money.

 

[cheapchips]

Did Frontier and Elite Dangerous start their development with AWS and so had plenty of time to build their solution around it?

P2P high bandwidth networking with AWS low bandwidth was planned into Elite from the Kickstarter (it was in the FAQ). From a business point of view, it a very clever and flexible solution, since with scales very well at low cost.

It took them a while of the live game to iron out kinks. When you move from their equivelent of quantum drive to normal space, it does matchmaking and creates a 'book keeping' instance on AWS. There were some issues with certain routers and conditions that caused huge pauses in this stage. It's mostly a few seconds to transition these days, but it did take years to resolve. Big events happening, like the recent station attacks, can cause the supercruise dropout to take longer. That's understandable though.

They also had issues with players that were grouped together in wings (party) not ending up in the same instance and messages not working. Think those are fixed these days. Again, issues that it took them a long time to figure out.

They still have problems with combat logging. It's not really an issue that bothers me, but it seems to bother more serious PvP types. It is a downside to P2P that's hard to fix fully. SC backers seem much more PvP orientated to me.

With P2P the number of physics objects and players SC aspires to would be an issue I'd have thought. I've have seen presentations on large scale P2P MMOs before, but these have been for the collision free variety.

Elite does do interesting solutions for specific things. The recent alien ships have swarms of mini fighters. Only a virtual swarm leader is networked, sharing broad position and status of the swarm. The swarm itself make local decisions based on deterministic rules, so each player still sees the same thing.

Wouldn't help with a collection of stacked boxes in SC though.

 

[MfA]

If you want to have Eve scale battles you need client server. Migrating zones in a single tick is easier with your own server racks with your own infiniband network and your own SSD's all with predictable performance, in the cloud everything is less predictable.

 

[CSI PC]

P2P high bandwidth networking with AWS low bandwidth was planned into Elite from the Kickstarter (it was in the FAQ). From a business point of view, it a very clever and flexible solution, since with scales very well at low cost.

It took them a while of the live game to iron out kinks. When you move from their equivelent of quantum drive to normal space, it does matchmaking and creates a 'book keeping' instance on AWS. There were some issues with certain routers and conditions that caused huge pauses in this stage. It's mostly a few seconds to transition these days, but it did take years to resolve. Big events happening, like the recent station attacks, can cause the supercruise dropout to take longer. That's understandable though.

They also had issues with players that were grouped together in wings (party) not ending up in the same instance and messages not working. Think those are fixed these days. Again, issues that it took them a long time to figure out.

They still have problems with combat logging. It's not really an issue that bothers me, but it seems to bother more serious PvP types. It is a downside to P2P that's hard to fix fully. SC backers seem much more PvP orientated to me.

With P2P the number of physics objects and players SC aspires to would be an issue I'd have thought. I've have seen presentations on large scale P2P MMOs before, but these have been for the collision free variety.

Elite does do interesting solutions for specific things. The recent alien ships have swarms of mini fighters. Only a virtual swarm leader is networked, sharing broad position and status of the swarm. The swarm itself make local decisions based on deterministic rules, so each player still sees the same thing.

Wouldn't help with a collection of stacked boxes in SC though.

Thanks for the info.
Yeah, although I remember Frontier did a high user count persistent test early 2017 in a PvP "arena" setup involving a lot of people for interracted instances, was over 1000 people.
I just think SC may need to go same type of solution as Frontier with Elite Dangerous, who have a lot more time to work with AWS and their own peer-to-peer integration.

 

[cheapchips]

Yeah, although I remember Frontier did a high user count persistent test early 2017 in a PvP "arena" setup involving a lot of people for interracted instances, was over 1000 people.

Don't remember that specific test. The highest in a live instance was just over a hundred I think. It's normally capped lower, but one of the expeditions tricked the cap higher.

I don't think P2P would work for what SC wants to achieve. You need it all happening server side with dumb clients just rendering stuff. We'll see what they actually release through this year in terms of improvement.

 

[eastmen]

played a little tonight , performance is rough , i only made it into my ship and tried to take off and crashed lool. I will try a bit more this weekend while i have time. I will say that the game is really dark, going to have to up the brightness

 

[CSI PC]

Don't remember that specific test. The highest in a live instance was just over a hundred I think. It's normally capped lower, but one of the expeditions tricked the cap higher.

I don't think P2P would work for what SC wants to achieve. You need it all happening server side with dumb clients just rendering stuff. We'll see what they actually release through this year in terms of improvement.

I will see if I can find it when I have time, may be some references to it out there.
Eve Online is probably one of the most complex games out there for server side requirements and is still also peer-to-peer.
I just cannot see how this will work well without it being a mix of both when using AWS.

 

[Jupiter]

CCP has the huge advantage that the amount of data is low compared to other MMOs. There are no fast movement changes, there are hardly visible items, the data is relatively robust in terms of latency. Whether you have 20ms or 200ms in Eve Online has almost zero impact. In a first person environment like Star Citizen 200ms are unacceptable. CCP also has the same problem as everyone else who has been in business for a long time. They are in a technological dead end. Their technique only scales within the framework of how they can buy better hardware. And for a change in technology they have already invested too much into the existing one.

But that is also a general problem. The existing technology is now so well documented that almost everyone only knows what they are familiar with. Much of the stuff comes from the days when all believed Moore's Law would solve the scaling problems. But if the goals and visions of a project go beyond what is already state of the art one also has to be ready to break new ground and not be allowed to re-program what already has known borders.

 

[CSI PC]

CCP has the huge advantage that the amount of data is low compared to other MMOs. There are no fast movement changes, there are hardly visible items, the data is relatively robust in terms of latency. Whether you have 20ms or 200ms in Eve Online has almost zero impact. In a first person environment like Star Citizen 200ms are unacceptable. CCP also has the same problem as everyone else who has been in business for a long time. They are in a technological dead end. Their technique only scales within the framework of how they can buy better hardware. And for a change in technology they have already invested too much into the existing one.

But that is also a general problem. The existing technology is now so well documented that almost everyone only knows what they are familiar with. Much of the stuff comes from the days when all believed Moore's Law would solve the scaling problems. But if the goals and visions of a project go beyond what is already state of the art one also has to be ready to break new ground and not be allowed to re-program what already has known borders.

You missed the point, visible-fast moving points/etc fall under peer-to-peer when looking at AWS in an efficient solution, which is what Frontier did with Elite Dangerous.
The context I thought I was responding to was other aspects of algorithms and database/persistent variables across instances and users, and here Eve is very complex while also being peer-to-peer.

 

[MfA]

Eve is not peer to peer.

 

[CSI PC]

Eve is not peer to peer.

From what I understand Eve is a hybrid that also includes peer-to-peer/mesh in my context.

Edit:
Looking now some sources suggest it is same concept as Frontier with Elite Dangerous and combination of server hosted and peer-to-peer *shrug*.

 

[Jupiter]

I made a thead in Star Citizen forums some people are getting angry about this. ^^

Is it still possible to produce a large MMO with the existing code base?
https://www.robertsspaceindustries....-still-possible-to-produce-a-large-mmo-with-t