servers = n-1

endar

For the longest time, the amount of game servers you could run on a single physical box was n-1, where n=number of cores.
Quad core = 3
Hex core = 5

But now, with the change to how server.exe handles wait states, cpu usage has dropped! I just had a look at my current usage. I have 3 servers running
#1 is using about 22%
#2 is using about 15%
#3 is using about 15%

System idle is hanging quite happily around the 48-52% mark, with very occasional spikes. With this in mind, I am going to try launching a 4th server at the 18 player mark, in the hope that it only adds another 15%. I'll monitor and post feedback. Im curious if anyone else is considering doing the same.

ScardyBob

We haven't for the Team156 servers, though I am a bit curious if the latest server-side changes have invalidated some of the previous rules-of-thumb for dedicated servers. For example, tying a server to a core may not matter anymore. Also, I think as long as all four servers aren't full at the same time, you'd have no issues running n number of servers.

falc

<!--quoteo(post=1972961:date=Sep 6 2012, 08:46 PM:name=ScardyBob)--><div class='quotetop'>QUOTE (ScardyBob @ Sep 6 2012, 08:46 PM) <a href="index.php?act=findpost&pid=1972961"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->as long as all four servers aren't full at the same time, you'd have no issues running n number of servers.<!--QuoteEnd--></div><!--QuoteEEnd-->

Thats not really the way a server op should calculate things like this. ;)

endar

And I never would!

Those numbers I was giving were the following:
24/24 players: 22%
18/18 players: 15%
13/18 players: 15%

It would be stupid to overprovision server capacity for this game, at this stage, since quite often (especially in Australia at least), all servers are full and there are more players than server slots.

It will only get more skewed when the game actually releases, it would be a disaster to actually reach 100% cpu usage, and force tickrates to suffer due to having too many servers.

Last night I never actually saw all four servers fill up, had three of them full and one half full.

And as to locking a server to a single core, I still debate whether that any had benefit (except to prevent a bug from consuming too much cpu)>

falc

<!--quoteo(post=1973130:date=Sep 7 2012, 12:57 AM:name=endar)--><div class='quotetop'>QUOTE (endar @ Sep 7 2012, 12:57 AM) <a href="index.php?act=findpost&pid=1973130"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->And as to locking a server to a single core, I still debate whether that any had benefit (except to prevent a bug from consuming too much cpu)><!--QuoteEnd--></div><!--QuoteEEnd-->

As soon as the main thread uses almost the complete capacity of "the" one core and the other threads are also forced to share the same core, the server is crippled performance-wise, which is pretty logical. I have tested this many builds ago ...

If the core has enough capacity to handle all threads of the server without timing issues, it eventually could be beneficial since it could prevent the scheduler from wasting performance. But this is highly theoretical, since it would require perfect timing by the threads and the os-environment - and is also almost impossible.

ScardyBob

<!--quoteo(post=1972963:date=Sep 6 2012, 11:52 AM:name=falc)--><div class='quotetop'>QUOTE (falc @ Sep 6 2012, 11:52 AM) <a href="index.php?act=findpost&pid=1972963"><{POST_SNAPBACK}></a></div><div class='quotemain'><!--quotec-->Thats not really the way a server op should calculate things like this. ;)<!--QuoteEnd--></div><!--QuoteEEnd-->
I was thinking more of 3 public and 1 private server, such that you know the private server wouldn't be populated most of the time.

endar

Had some minor lag tonight due to running four servers. I guess it was bound to happen, when they are running a standard game, the 18 players seems to stay around 20. But I was in a 80+ minute slugfest, and it hit the usual 28-29, which caused 100% overall.

Back to 3 servers.