from my server hosting experience; in dxmp, theres something very odd, on server side mostly. client side is ok, except from the nasty gpf bugs.
on server side: it looks like the game is checking all actors (active networking objects?) in game and updates their state at certain rate. dxmp has a lot more actors than unreal tournament. in ut, you have players, their weapons and weapons lying around (maybe some more? didnt play ut so much). in dxmp you have players, their weapons, their granades, items, even multiple of them. if you carry 5 bios, 5 meds, 5 lockpicks, multitools, and 3 of each granades + 3 weapons you have total of 32 actors that are being updated every x msec. in ut, considering you can carry max 9 weapons (less or more, dunno?), there is already big difference here. plus you have to consider that dxmp has a lot more actors in the world, all that items laying around + many weapon racks where are a lot of weapons.
thats why i found very interesting thing to improve server performance, when more players are playing; try to reduce number of actors in world. put only 2x2 weapon racks for TDM map, try to avoid putting in useless shit like crates and actors that have no use in multiplayer.
to test performance of server, join server with any map you like and use command: "inject userflag 1" ("inject userflag 0" disables messages). you will see whats the tickrate of server, how much time server takes for actor processing and how much for net processing (in msec). in dxmp is odd that most of processing takes net. now i can only assume, since its all related to number of actors in game, that net has special processing for all actors and there must be smth going very very wrong.
now to test my idea, try checking inject userflag on map like DXMP_Cmd which has quite a lot of actors and then make a simple map, a box and put in only playerstart, nothing else. and do inject userflag then
you will notice that your custom map will have server side processing for actors and net virtually zero. the values you get depends on servers side CPU speed.
so what ive learned; the best method in dxmp to achieve best performance for more players is simply the brute strenght and speed of CPU. faster the CPU is, more players your server will hold without increasing pings too much and giving lag.
thats why i still hold the record of 32 players (i think sr has screenshot of that) in server and pings werent 500 or more. the cpu was E4300 OCed at 3ghz with 333FSB.
i know this is not a lot, but theres coming even better performance soon. a friend that is offering dxmphosting.com servers is building a new server that will be shipped to germany for dxmp purposes. the CPU there is E8300 engineering sample which OCs like mad. hes promising to run it at 4GHz (at FSB 500!!!!). for all those that dont know this CPU, it has 6 megs of L2 cache (E4300 has only 2 megs). the performance compared to my OCed E4300 which i had are around 80% higher!!! if all goes well, you will soon see a new 32 slot server online that will be able to trully hold 32 players without any kind of lag and only for little highered pings.
