We would like to add multiplayer support for our game, with dedicated servers that we would manage.
The replication code works OK and the game works on a LAN.
We are much too small to write the master server and back-end services on our own and would like to find tools, services and libraries to help us.
We are looking for the usual online subsystem features (matchmaking, friends list, social features, leaderboard, vocal chat…)
The game is real-time (not turn based) and requires low latency.
Among the following options, how would you rate the amount the integration work needed ?
GameLift or PlayFab , perhaps with EOS or Steam ? (unclear if EOS or Steam is still required when using those?)
EOS with the Redpoint plugin + Agones for container spawning / autoscaling
(I like the idea of “just” requiring K8S and be able to choose any cloud provider that has support for it)
(perhaps some other solution ? Chilliconnect ? Zeuz ? SmartfoxServer ? )
Also, the pricing is complex to determine… Are there solutions known to be much more (or less) expensive than others? (I understand it may depend on monthly active and concurrent users, though)
I have read that some people find EOS too young to be used, but other are seemingly happy with it. The idea of targeting all plateform user bases with the same online subsystem looks extremely cool.
We can do C++ and a Linux port of the dedicated server is something we are willing to do.
Thanks in advance for any pointers to help shedding some light on this complex topic!!
Just a question more than an answer…
If you are a small company and or team and have no intention to expand thus requiring physical hardware,
Why not stick to the listen server method
Or implement a custom Peer to Peer approach?
Regardless, your own dedicated server at a reputable place (think IBM) runs about $170 monthly to $2000+ depending on the server components and the bandwidth.
Going through non name brand places can reduce that cost to about $50 monthly for the lower end.
On the other hand you pretty much need a Linux or windows server admin on a paid salary to maintain it, so the cost of that is around $100k yearly either way.
(Mind you that it’s not something that anyone can just do, especially to keep games running. Someone in your team should be dedicated to it and either know it, or learn how over time).
As far as implementation work, it’s a long process whichever way you go, which is why a custom p2p is a smarter solution even if you have to actyally hire a network engineer to make it happen - it would lilely cost you less in the long term too…
The game has stringent latency requirements and low CPU usage (server-side). We would like to supply the best player experience, and this is why the plan is to supply dedicated servers. We expect to be able to run a lot of connections on a single vCore.
Also, regarding server admin, I am more inclined to work in a SaaS vs PaaS way, where the server will be containerized, so that the hardware details are handled by some other party, namely a cloud provider (big one or smaller player – we are still in an exploratory phase)
The idea is to trade upfront fees for pay-as-we-go where the load in terms of # players would auto-scale the server instances.
Cost wise, you’d probably be better off creating a dedicated server instance mirror image and adding in new servers as needed.
Most service/usage based solutions are VMs, which don’t necessarily run all that well.
Azure would be a decent contender on that, but the cost can quickly become astronomical above 32gb of ram.
Also, regarding the lag, p2p and listen server handle really well if the point of the game is playing with friends near you.
If it’s like an MMO, than no - and there’s a bunch of pros/cons to consider when choosing the service or server provider too… (including azure as it will tie your VM to to central location).
Thank you. It’s not an MMO, on the server side it’s more like a MOBA : sessions are separate and not long-lasting. So, no need for a central location.
I will give PlayFab a test, because the cost of writing software that automatically manages dedicated servers is a bit too much for a small team. It will have to be balanced with the (potentially big) PlayFab costs.
This first proof of concept will also check whether latency between the clients and the container-based dedicated server is acceptable or not.
I love the Kube as much as the next person, but there is a real cost in latency for workloads that aren’t given dedicated cores. If you have a single container that runs on one CPU core, and serves many connections, that can still work, but if you think you’d run many sub-core-size containers sharing an allocated core, I’d recommend against that.
I don’t have experience with GameLift or PlayFab, so I can’t help there. If latency is important, sitting on top of globally well distributed data centers, and having a PoP in each major hub, seems important, which would indicate you’d want Amazon or Google or just maybe Azure clouds, though. They each have auto-scaling solutions that work OK, so you might want to look at those too. But, again, most infrastructure intended to run containers tends to assume web-like traffic, which ends up with noticeable latency jitter.
What drew me towards PlayFab is the fact that it’s Microsoft-owned and runs on top of Azure. Plus I can customize the “instance size” vs “number of server processes per instance” (hence, hopefully, core affinity) and, above, all, I have discussed with people being happy with its latencies.
We are dealing with this soon enough so that we can backtrack on an alternative solution, perhaps involving bare metal.
You should swing by NFOServers.com and talk to Edge100x. They do/did have a program for developers (free machines)… bare metal dedicated boxes. At a minimum you might be able to hardware profile your game.
