Let's talk about 255/1000+ player on dedicated server... how? hardware?
by elvince · in Torque 3D Professional · 04/01/2010 (11:57 am) · 19 replies
Hi,
When you look at last year the main page of Torque, Garage Games were announcing dedicated server that can manage 250 player for Fps type and 1000+ for Click'n play style.
Let talk about that by thinking we are using the stock T3D:
What type of hardware you need to reach that? In term of CPu, memory etc...
I'm afraid as now this number of player per server is not put anymore on the T3D page. Does that mean this is no more the case? Did someone get to simulate the X number of player and have some "real life" figures?
I'm very interesting in people (or GG) that made some real test with the current T3D and have their feedback.
I understand we are moving from 250 to 1000 between the 2 game style because you have less collision to manage? Is it the case or I miss other major differences? what make a game style taking 4 times more "power" than another style?
In term of bandwidth, I read previous statement on TGE, TGEA?, that a player is consuming about 8ko/s. Is it still the case in Stock T3D?
I get those figures from this statement:
Maybe some people may ask why I'm looking for that:
I choose this game engine because of his good netcode/network performance and for other reasons, but this one was important.
I'm working on the gameplay design with my team, and I don't want to make bad decision just because the server won't fit or I will never get the hardware needed to run my game at a reasonable cost... Having this information in my hand may also drive me in certain way to design my game.
If I can make those choice sooner, it will save me time & disappointment at the end.
Thanks !
When you look at last year the main page of Torque, Garage Games were announcing dedicated server that can manage 250 player for Fps type and 1000+ for Click'n play style.
Let talk about that by thinking we are using the stock T3D:
What type of hardware you need to reach that? In term of CPu, memory etc...
I'm afraid as now this number of player per server is not put anymore on the T3D page. Does that mean this is no more the case? Did someone get to simulate the X number of player and have some "real life" figures?
I'm very interesting in people (or GG) that made some real test with the current T3D and have their feedback.
I understand we are moving from 250 to 1000 between the 2 game style because you have less collision to manage? Is it the case or I miss other major differences? what make a game style taking 4 times more "power" than another style?
In term of bandwidth, I read previous statement on TGE, TGEA?, that a player is consuming about 8ko/s. Is it still the case in Stock T3D?
I get those figures from this statement:
Quote:
Upload Bandwidth equal to or greater than 0.5 Mbps allows up to 8 players
Upload Bandwidth equal to or greater than 1.0 Mbps allows up to 16 players
Upload Bandwidth equal to or greater than 2.0 Mbps allows up to 32 players
Maybe some people may ask why I'm looking for that:
I choose this game engine because of his good netcode/network performance and for other reasons, but this one was important.
I'm working on the gameplay design with my team, and I don't want to make bad decision just because the server won't fit or I will never get the hardware needed to run my game at a reasonable cost... Having this information in my hand may also drive me in certain way to design my game.
If I can make those choice sooner, it will save me time & disappointment at the end.
Thanks !
About the author
Recent Threads
#2
In fact, i just wanted to know from where we are starting From with stock t3d.
In GG highlighted this information during first sales, i suppose some test have been made and I will apreciate their feedback on this thread.
I m not expecting that they simulated a 250 game but maybe they test 150 and extrapolate to 250 with the performance figure they had.
It could be great to understand the hardware we need. Does multicore will be More efficient than multiple cpu? Etc...
I'm sadly not enough an expert to extrapolate a config from the source code.
In all case i Will try to create a dummy class to simulate player activity and see how the server react.
It could be great also if you share some result from your testing. I m thinking also about repopulation feedback that could be interesting.
04/02/2010 (5:19 am)
I agree that i need to take care about information transfers. In fact, i just wanted to know from where we are starting From with stock t3d.
In GG highlighted this information during first sales, i suppose some test have been made and I will apreciate their feedback on this thread.
I m not expecting that they simulated a 250 game but maybe they test 150 and extrapolate to 250 with the performance figure they had.
It could be great to understand the hardware we need. Does multicore will be More efficient than multiple cpu? Etc...
I'm sadly not enough an expert to extrapolate a config from the source code.
In all case i Will try to create a dummy class to simulate player activity and see how the server react.
It could be great also if you share some result from your testing. I m thinking also about repopulation feedback that could be interesting.
#3
now all things being UNequal (the old TGE engine did a lot less than the new T3D engine is doing, the connections were a a bit slower during that era :) , i was on dial up, and i had a slower computer )... on several occasions i managed to connect to games (hosted by regualr developers on their own personal computer systems) with as many as 32 players...
... and there was lil if any lag or latency issues, and you could shoot a player a loooong ways off, even when he was running... and hit him.
in short... me thinks that the net code in Torque is simply some of the best performing logic i'll ever get to play with, and realistically, i would suggest that you scale a test project up... from 8 to 16... to 32 players... then higher...
i think that's the only way to actually get a 'real feel' for what you can expect...
good luck
--Mike
04/02/2010 (5:56 am)
way back when... :) at the dawn of time, when garage games was in its infancy... so to speak, a few people built first person shooter levels based on the demo...now all things being UNequal (the old TGE engine did a lot less than the new T3D engine is doing, the connections were a a bit slower during that era :) , i was on dial up, and i had a slower computer )... on several occasions i managed to connect to games (hosted by regualr developers on their own personal computer systems) with as many as 32 players...
... and there was lil if any lag or latency issues, and you could shoot a player a loooong ways off, even when he was running... and hit him.
in short... me thinks that the net code in Torque is simply some of the best performing logic i'll ever get to play with, and realistically, i would suggest that you scale a test project up... from 8 to 16... to 32 players... then higher...
i think that's the only way to actually get a 'real feel' for what you can expect...
good luck
--Mike
#4
I know that there were some amazing results in games that used the very same networking layer - for example NCSoft's Auto Assault. I remember that it seemed exceptionally smooth and solid (I didn't yet know about Torque at the time).
I think planning for about 64 players using Torque out of the box on a decent (yet desktop type) server with bandwidth that does not cause a bottleneck is achievable. If you want more, you'll need to start customizing code.
While I've spent some time on this - creating new objects that use less bandwidth - I'm still in the dark in many aspects of the networking layer. Our game requires only about 64 players on one mission at a time, but they all will be frequenting the same places on the map, so I'm trying to prepare for the worst.
04/02/2010 (6:12 am)
I'll gladly share any results I have once I have any, really. I was thinking the tests further - perhaps not the player object but the gameconnection object (the client itself) should be used when creating a dummy test object.I know that there were some amazing results in games that used the very same networking layer - for example NCSoft's Auto Assault. I remember that it seemed exceptionally smooth and solid (I didn't yet know about Torque at the time).
I think planning for about 64 players using Torque out of the box on a decent (yet desktop type) server with bandwidth that does not cause a bottleneck is achievable. If you want more, you'll need to start customizing code.
While I've spent some time on this - creating new objects that use less bandwidth - I'm still in the dark in many aspects of the networking layer. Our game requires only about 64 players on one mission at a time, but they all will be frequenting the same places on the map, so I'm trying to prepare for the worst.
#5
I mean this was already the case with Tribes2. I think some improvement have been made since so I hope we can have more from a fresh install.
AS you mentioned, gameconnection may be the class to look at as it manage your connection to server.
I didn't know that Auto Assault was based on the same netcode. any other "known" game that use this netcode?
64 at the same place, that will be a good test :D I hope you won't find issues.
How many AI do you think you can add to the 64 players? I know that AI can be very consuming for servers, and this can lead to slow down the server.
04/02/2010 (10:27 am)
64 players with out of the box, that's ok but not so great.I mean this was already the case with Tribes2. I think some improvement have been made since so I hope we can have more from a fresh install.
AS you mentioned, gameconnection may be the class to look at as it manage your connection to server.
I didn't know that Auto Assault was based on the same netcode. any other "known" game that use this netcode?
64 at the same place, that will be a good test :D I hope you won't find issues.
How many AI do you think you can add to the 64 players? I know that AI can be very consuming for servers, and this can lead to slow down the server.
#6
As for the 64 players out of the box - it's just my guess. It could be more, but I'm pretty sure it's not less. But, I have not yet made any tests to back that up.
I'm not familiar with who had licensed the TNL before and after it became open source - just knew about Tribes and Auto Assault of the bigger names. Wikipedia's article on GarageGames might give you more information about that.
04/02/2010 (11:29 am)
AI will be fast if it does not think frequently, the think cycles are well optimized, pathfinding solutions are cached and most or all of the code is C++. I have not yet made any tests regarding this, but I'm close to accomplishing all these and give things a try. Still, we won't have an extreme amount of agents, so I probably will not push that to the limit.As for the 64 players out of the box - it's just my guess. It could be more, but I'm pretty sure it's not less. But, I have not yet made any tests to back that up.
I'm not familiar with who had licensed the TNL before and after it became open source - just knew about Tribes and Auto Assault of the bigger names. Wikipedia's article on GarageGames might give you more information about that.
#7
Wasn't tribes 2 even 128 player servers? And now we also have servers that have like ten times the raw power than we had back then.
I wonder how many players can be handled by your average single core cpu dedicated server with a dedicated 100Mb line. In best case, like without accounting for AI thinking, just by managing online players.
04/02/2010 (12:12 pm)
(I'm gonna drop my poorly worded topic for this more prolific one)Wasn't tribes 2 even 128 player servers? And now we also have servers that have like ten times the raw power than we had back then.
I wonder how many players can be handled by your average single core cpu dedicated server with a dedicated 100Mb line. In best case, like without accounting for AI thinking, just by managing online players.
#8
We have a couple of dedicated 100 Mb lines ready for launch (which I'm really happy about), so I too am curious what the numbers will be. I'll do my best to prepare some sensible report on that. That will most likely happen sometime at the end of the summer.
I could however put up a vanilla 1.0.1 dedicated on one of our servers. If there were enough people to give it a try, we could gather valuable data.
04/02/2010 (12:18 pm)
I don't know the exact player caps for Tribes. You're right, it could be a lot more.We have a couple of dedicated 100 Mb lines ready for launch (which I'm really happy about), so I too am curious what the numbers will be. I'll do my best to prepare some sensible report on that. That will most likely happen sometime at the end of the summer.
I could however put up a vanilla 1.0.1 dedicated on one of our servers. If there were enough people to give it a try, we could gather valuable data.
#9
konrad,I m ready for the basic test. I remembered someone else post something about doing a real test but nô news since one month.
I hope we can reach 100+ person to do this test.
If gg can advertise on it in the blog, we may be in position to get a lot people available at the same day.
04/02/2010 (12:35 pm)
I checked and tribes was 32vs32 game at maximum. That's mean we can clearly expect far more as technology has evolved and raw Power has also taise up a lot. konrad,I m ready for the basic test. I remembered someone else post something about doing a real test but nô news since one month.
I hope we can reach 100+ person to do this test.
If gg can advertise on it in the blog, we may be in position to get a lot people available at the same day.
#10
04/02/2010 (12:43 pm)
Sounds good. The soonest I can set this up is next Tuesday though. I'll leave it like that for a while, so anyone can test it - even alone. We'll see if we can get enough people to do a stress test.
#11
I can participate if you guys want, as long as i have all the informations :)
04/02/2010 (12:57 pm)
i'm poking around to create some kind of "fake client" to simulate connections to the server, not having much success for now...I can participate if you guys want, as long as i have all the informations :)
#12
04/02/2010 (9:02 pm)
Count me in, send an email with instructions or post it up here somewhere if you would.
#13
Any update on this server? Have you been able to put it online?
Thanks,
04/10/2010 (9:44 am)
Hi Konrad,Any update on this server? Have you been able to put it online?
Thanks,
#14
We were hoping to launch a new service by last Tuesday, but we had to delay the launch and postpone it to this Monday. I'll get this done as soon as that stuff is behind me. Will let you know when it's up - most likely on Tuesday.
04/10/2010 (9:51 am)
Forgive me, I haven't been able to get there yet. We were hoping to launch a new service by last Tuesday, but we had to delay the launch and postpone it to this Monday. I'll get this done as soon as that stuff is behind me. Will let you know when it's up - most likely on Tuesday.
#15
04/10/2010 (10:05 am)
ok no pb & no stress around this :D
#16
09/05/2010 (10:17 pm)
Sooo i guess this kinda didn't happen? Any news? 
#17
Your going to have to learn C++ and do a lot of work to make this happen.
09/06/2010 (2:15 am)
Hardware is not your answer BTW, I have run torque on some of the biggest machines you can purchase and it does not help.Your going to have to learn C++ and do a lot of work to make this happen.
#18
09/06/2010 (2:39 am)
Yeah but this was a curiousity thing, to my knowledge nobody benchmarked the example fps base to see how many simultaneous players it can handle running and shooting around.
#19
What were your result on this machine, How many player did you succeed to reach without lag?
What were the specs? the bandwidth?
Thanks,
@Kyrah,
That was the point of this thread to benchmark it to have a clear view of the performance outside the box.
09/06/2010 (6:19 am)
@Chris,What were your result on this machine, How many player did you succeed to reach without lag?
What were the specs? the bandwidth?
Thanks,
@Kyrah,
That was the point of this thread to benchmark it to have a clear view of the performance outside the box.
Associate Konrad Kiss
Bitgap Games
Another question is where the players appear. Are they scattered all over a huge world? How likely is it that one client will ghost most or all other clients?
For testing I'd create a very simple geometry for the players, so the tests would not be disturbed by having too many draw calls when there are too many players on the screen.
There are many things to optimize in Torque's built in objects. You should always check what information is sent over the wire. For example, if you set the player's energy, it will dirty the appropriate flag and trigger a refresh of the value on the client. This is cool, but it is done even if the energy has the same (new) value as its previous value.
Another example would be projectiles. If you create an rpg, or anything that makes projectiles hit a target based on player skills and scores and not on actually aiming with the mouse, then you don't need to have every client ghost every projectile all through their flight course.
I'm pretty sure, that with the right optimizations, 1000 players can be achieved. Perhaps even more. But I doubt it would be possible with the fps nature of the built in objects and demos that come with Torque. You'd need to fine-tune your code to your game to get the best results.