Dedicated servers and multicore support
silentsushix3
Join Date: 2013-01-09 Member: 178337Members
So whats going on with the future of ns2! Will we finally see multi core support or is there any help about boosting performance.
My processor is at 7% with a full 24 man server but the first core is like 96% and may lag at times? How can i fix this or what is going to be done to fix this?
My processor is at 7% with a full 24 man server but the first core is like 96% and may lag at times? How can i fix this or what is going to be done to fix this?
Comments
Until then your only real option is to throw fast hardware at it.
My processor is at 7% with a full 24 man server but the first core is like 96% and may lag at times? How can i fix this or what is going to be done to fix this?<!--QuoteEnd--></div><!--QuoteEEnd-->
As for now. Disable hyper threadening and use any core you would like BECIDE the first one (core 0). Use high priority on that core you choose for your game server.
While I agree it would be great for more companies to utilize full multicore support in there games it could open up a door server providers don't want to see. Very intense and unoptimized games would probably become more common since they could take advantage of multiple cores but that wouldn't make them any cheaper to run, therefor hurting the consumer.
I think they need to be focused on optimizing their game and not worrying about what sort of BS some other company would do.
I think they need to be focused on optimizing their game and not worrying about what sort of BS some other company would do.<!--QuoteEnd--></div><!--QuoteEEnd-->
I was speaking more in general there, the more people that start doing it the better chance it could become standard. I agree that focusing on optimizing would be better then multicore support.
Agreed with you 100%.
<!--quoteo(post=2059352:date=Jan 11 2013, 07:02 PM:name=kraze)--><div class='quotetop'>QUOTE (kraze @ Jan 11 2013, 07:02 PM) <a href="index.php?act=findpost&pid=2059352"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->I was speaking more in general there, the more people that start doing it the better chance it could become standard. I agree that focusing on optimizing would be better then multicore support.<!--QuoteEnd--></div><!--QuoteEEnd-->
Yes. Absolutely.
Bottom line is, most people will not dedicate all cores in a machine for ns2. It needs to be further optimized for 1-2 thread operation.
All the client side games support this nowadays. Servers mostly aswell, and they do support more then 1 core you can set the affinety therby setting it to still run on a single core.
<!--quoteo(post=2058926:date=Jan 10 2013, 08:01 PM:name=kraze)--><div class='quotetop'>QUOTE (kraze @ Jan 10 2013, 08:01 PM) <a href="index.php?act=findpost&pid=2058926"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->While I agree it would be great for more companies to utilize full multicore support in there games it could open up a door server providers don't want to see. Very intense and unoptimized games would probably become more common since they could take advantage of multiple cores but that wouldn't make them any cheaper to run, therefor hurting the consumer.<!--QuoteEnd--></div><!--QuoteEEnd-->
Server providers won't mind. Nowadays most servers are virutal servers which most of the time use shared resources. They would rather have a 2 core server run both it cores at 50% percent then 1 core @ 100%. Escpecially VPS servers which share recourses.
While it is harder to implement multicore support I do see why the NS team chose to not implement it at the start. It would have meant a lot more work. The servers could indeed handle more players easily but this game is not a "16vs16 game". The maps are not made for that and the teambalance isn't aswell. I played a hacked 32 player server once and believe me that was no fun. I tend to avoid 24player games for the same reason.
Now you could say they could make larger maps, but that is where te limitations of the spark engine khick in. With a visibility of max 45 meters larger maps are kinda out of the question.
So while I do believe one day the NS team will improve the server performance I still doubt their main focus will be multicore support.
I'm thinking multicore utilization could possibly remedy all this? But again I do understand this is potentially a huge time and money investment by UWE but... perhaps the payoff is worth it?
Food for thought.
I'm thinking multicore utilization could possibly remedy all this? But again I do understand this is potentially a huge time and money investment by UWE but... perhaps the payoff is worth it?
Food for thought.<!--QuoteEnd--></div><!--QuoteEEnd-->
You can, I'm currently running three 20 slot servers on a E5-2690 in a virtual environment(six dedicated HT cores), I've also tested with E3-1270's. All servers run just fine.
I wish there was some sort of way to run server side (client side would be nice too) benchmarks that could help to test what and where we are having issues. In CSS for example I would just fill a server with bots to simulate load and test out different scenarios to then see how mods ect effect the performance and test for memory leaks.
Currently from what I have seen the largest bottleneck is either in memory bandwidth/size or in IO (possibly due to extensive logging combined with lack of append feature).
Example right now: (both full)
18 player ns2 server 646,144k and 16cpu (memory usage here is lowest I have seen in days/weeks usually over 1gb)
40 player zombie server 479,972k and 19cpu
\Steam\steamapps\common\natural selection 2\utils\PerfAnalyzer
I'm not entirely sure which version of Python you'd need though (or how it actually works :P)