Hi.

Im working on a multiplayer game, The game is mostly complete and were at a stage were were doing a weekly playtest sessions with some friends etc, soon I want to start getting more people to help with playtesting (or just to play in general).

I am currently hosting it on some relatively cheap VM but the response times doesn't feel great (it has realtime RTS like combat), something I noted was that sometimes actions register almost immediately and sometimes it takes up to what feels like half a second which makes me believe the VM is at fault due to how it divides CPU time). I want to move to a dedicated server to see if its going to be better (I'm almost certain it would be since i've seen this happen in the past as well, although I was hoping that VM's have perhaps become better in that regard than they were 10 years ago, or perhaps this particular provider just oversells they're resources).

Now as I started looking for a dedicated server provider I noticed that both google and amazon have some kind of “scaling game server platform” solution were they are promising performance and the best possible latency from all over the world, and also a whole bunch of other features that I don't care about (or at least not at this point…)

My question is. Can anyone give me some insight into google/amazons solutions? or otherwise recommend a good provider for getting a dedicated server for gaming (we'll I'm pretty sure any of the big providers would be fine but maybe someone has some more insights) or even a provider that offers good VM's that are reputable as good for hosting Realtime game servers.

What is your budget? The main reason to use “scaling game servers” is when your servers measure in dozens or more of machines.

It used to be that the really small Amazon instances would sometimes go away for a whole second, and I think that's still happening, but the “medium size” instances behave a fair bit better. Go for 4 cores or more and you'll see much less mystery lag in my experience.

A simple way to measure the latency performance of a virtual server is to write a program that keeps running “get the current time, sleep for 100 microseconds, measure the time again,” and measure how much longer the time interval is than the expected 100+epsilon microseconds. You might want to log a message when the time is LESS than 100 microseconds, or when it's more than some delta, such as 3 milliseconds. Then run one of these program instances per (virtual) core on the machine, and see how many events happen, with how much latency.

Then run on a normal machine under your control, and compare. OS-es aren't perfect, so SOME small bumps are expected no matter what. The main question is how high up the instance tier you need to go before you get good enough performance.

Also, the Amazon ARM-based server instances are pretty great in my experience. They have twice the raw throughput of the corresponding Xeon instances, for about 70% of the cost! Assuming your server runs on Linux, you might try re-building for those instances and try on something like a m6g.2xlarge or c6g.2xlarge instance.

OK, per your recommendation I got a m6g.2xlarge instance from amazon, I've spent the last 12 hours getting my server and related tools running on ARM, little snags here and there but overall not too pain full, I just finished everything and I'm extremely happy with the performance, the online gameplay became silky smooth. Thanks a lot!! once again your advice helped me out a lot.

Glad to hear it. Also, thanks for reporting your experience on the m6g.2xlarge instance type! This is useful to many people on the boards.

Indeed hplus0603! I for one would find that info quite useful to planning out hosting for my own multiplayer game. Especially around any up/down scaling as demand changes and any implementation challenges with the initial setup of a single host.