I'm Curious. The Lag Is Caused By What?
ViPr
Resident naysayer Join Date: 2002-10-17 Member: 1515Members
Resident naysayer Join Date: 2002-10-17 Member: 1515Members
Comments
Also, excessive turret farming lags the server good and hard...
<snip>
Just wanted to update you really quickly with a recent development. I've made a big performance optimization, such that all spawn entities are tracked independently of HL, and are deleted when the map loads. This means there are now 96 less entities on all NS maps! With other improvements from the mappers and some other code changes, this brings some maps below the 200 range (!), and most of the other maps below 300. This should make a HUGE difference in server performance in 1.1 and beyond.
I won't say that it's at CS performance level yet, but it should be a lot better.
P.S. ...and no, please don't ask for an ETA on it, it'll be ready as soon as it's ready.
</snip>
A lot of work that the server has to do depends on the # of entities on the map. Less entities = faster server.
That's why some of the more complex maps (like hera) can cause more lag than the simpler maps.
i was hoping the lag would be reduced without removing entities from maps and making them more crappy and simple.
btw i seen weird things. many transparent surfaces just pop up in view when entering a room. that means they are server side. they are not supposed to be because they don't move or break or anything. also many windows can be jumped straight through before you get replaced back in front of them and your movement lags when you are touching some windows meaning they are server side when they shouldn't be coz they are not breakable. same thing with stepping on resource nozzles.
The server essentially runs a full HL client to keep track of who can see/hear what, hivesight, motion tracking, presumably all the structures. This is why the dedicated server download is so bloody huge - if you look, it has all the models, sounds, etc. A normal HL dedicated server is quite tiny by comparison, and the memory/cpu usages of the NS server versus a standard one differ accordingly.
My guess is that Flay is managing to find various entities (such as the player spawns) that don't need to be kept track of by the server as it's updating every X frames, so he's giving them over to the client or finding a way to just get rid of them (I'm still not clear on what exactly is being done with the spawn entities).
Spawn points are not kept track of nor are they updated. They only matter on the server side and will never congest the bandwidth. Too many spawn points will simply make the map load longer, but only by a millisecond. In the communication between the HL client and the HL server, no update is given to info_player_start entities. The only update is given as when and where the player does spawn.
The only possibilities are the entities that actually can interact with the players. My best guess? Something to do with the buildings, as they're treated as monsters, meaning they do get updated (ie: sound, animation sequence, health) and as a result generate large amounts of bandwidth. Even the angle at which a single turret faces is conveyed to the client.
We have to understand that the HL client sees very few entities, but rather sees the effect of entities. So obviously a static armoury would require less bandwidth than say a functioning RT.
Flayra already knows what bugs are causing the lag, and he's busy fixing them.
My guess is that he forgot his welder one afternoon while coming back to his computer from the bathroom and a sneaky gorge webbed him up good, so he can't work very fast. This explains why 1.1 isn't coming out as fast as we'd all like and why he hasn't posted in this topic yet. <!--emo&:)--><img src='http://www.unknownworlds.com/forums/html/emoticons/smile.gif' border='0' style='vertical-align:middle' alt='smile.gif'><!--endemo-->
erm, what?