Overloading the server
Posted: Mon Feb 13, 2017 2:19 am
Hi guys,
I can't seem to find anything quite relevant to what I'm seeing, but I was just curious if anyone was aware of, or had an idea if maybe a fix was going to happen for an issue. The long / short is that we've noticed it's very possible to queue up a huge amount of work that can overload the dedicated server, which is turn causes it stop communicating with the clients until it catches up (Server not responding). If you attempt to drop something like a blueprint that has 5,000 items in it, the server CPU will peg at 100% and if it stays there too long drops all of the players out of the game. Once the "work" is done and the cpu goes back down to normal, everyone can rejoin the game again. The obvious answers to the problem itself is "don't do that" and "look for a way to increase the server timeout wait period", but curious if there was a way to either prioritize client connections or maybe move it off into a separate thread so that even if the server is completely loaded down with work, people don't get dropped out of the game (Maybe it's overloading the client side as well, I'm not sure). I'm seeing this with a dedicated server on an i7 and a lan client with an i7 that's even faster than the dedicated server, so I'd imagine it's probably more frequent for lower end hardware.
I can't seem to find anything quite relevant to what I'm seeing, but I was just curious if anyone was aware of, or had an idea if maybe a fix was going to happen for an issue. The long / short is that we've noticed it's very possible to queue up a huge amount of work that can overload the dedicated server, which is turn causes it stop communicating with the clients until it catches up (Server not responding). If you attempt to drop something like a blueprint that has 5,000 items in it, the server CPU will peg at 100% and if it stays there too long drops all of the players out of the game. Once the "work" is done and the cpu goes back down to normal, everyone can rejoin the game again. The obvious answers to the problem itself is "don't do that" and "look for a way to increase the server timeout wait period", but curious if there was a way to either prioritize client connections or maybe move it off into a separate thread so that even if the server is completely loaded down with work, people don't get dropped out of the game (Maybe it's overloading the client side as well, I'm not sure). I'm seeing this with a dedicated server on an i7 and a lan client with an i7 that's even faster than the dedicated server, so I'd imagine it's probably more frequent for lower end hardware.