[REL] dedicated server for Windows with no GPU
-
I guess someone from BMS should tell them the obvious……
-
-
Hi folks,
our server instance has started crashing when exiting from the 3d world (hitting ESC e). Any ideas what could be causing this? It’s kind of a bad problem because now we’re unable to save any campaign progress locally on the server once the server exits the mission.
All the best & thanks in advance,
Uwe
EDIT: Managed to fix the issue by making sure the remote sound would be played locally over the rdp connection (missing sound device, I guess).
-
Why are you stopping the server ? Ours stays in 3d until I am forced to reboot BMS (once a week or if the server jumps out of 3d)
I usually pause ours after we have finished. And do a quick save “Alt C : Q”
-
the server’s only running for campaign flights involving humans, that’s why we’re stopping the server (and saving the game) after MP sessions. The win7 VM gets rebooted daily, too.
Thanks for the quicksave keyboard shortcut, I guess that could come in handy.
Cheers, Uwe
-
Out of curiosity - why are you rebooting the VM at all?
sh
-
Out of curiosity - why are you rebooting the VM at all?
sh
Old habits die hard I guess.
Uwe
-
Just a quick update. x64 crashes instantly on launch, but x86 shows a splash screen, then crashes on initialization.
Edit- Here is the crash log. http://pastebin.com/txaLsX3d
Apologies for the slow response.
Set your console session’s desktop to 32-bit color with at least the resolution you’re using in BMS, and it has to be at least 1024x768 always.
Force 32-bit color when connecting through RDP.
Old habits die hard I guess.
Uwe
Have it whichever way you prefer, but be aware – Windows NT is as stable as Linux, especially after all these years. This isn’t windows 95->ME.
-
I’ve removed the auto-reboot, but lately we’ve been suffering abysmal fps both in BMS 2d and 3d when running the server on kvm / virsh. CPU load is only around 20% according to the gauges, the underlying centos server is idle, too (by that I mean no other tasks are running that migt have an adverse effect on BMS performance).
I increased BMS RAM to 8GB (out of 16 total), but that hasn’t helped, either. 8(
Uwe
-
for my information , is there any good side to run this server on the same machine my client run on ? (cpu is i7 , bunch of ram , win7 64 etc).
i’m thinking about better cpu usage, maybe improved client FPS in the case of a large TE is played ?
-
I’ve removed the auto-reboot, but lately we’ve been suffering abysmal fps both in BMS 2d and 3d when running the server on kvm / virsh. CPU load is only around 20% according to the gauges, the underlying centos server is idle, too (by that I mean no other tasks are running that migt have an adverse effect on BMS performance).
I increased BMS RAM to 8GB (out of 16 total), but that hasn’t helped, either. 8(
Uwe
Look into cpu type (should be host/native) and cpu topology (should be just like the real CPU). Also minimize the server window regardless of whether you disconnect from RDP.
for my information , is there any good side to run this server on the same machine my client run on ? (cpu is i7 , bunch of ram , win7 64 etc).
i’m thinking about better cpu usage, maybe improved client FPS in the case of a large TE is played ?
You can run both on 3 CPU affinity threads with little overlap. That’ll offload some processing. You’re likely right.
-
I’ve removed the auto-reboot, but lately we’ve been suffering abysmal fps both in BMS 2d and 3d when running the server on kvm / virsh. CPU load is only around 20% according to the gauges, the underlying centos server is idle, too (by that I mean no other tasks are running that migt have an adverse effect on BMS performance).
I increased BMS RAM to 8GB (out of 16 total), but that hasn’t helped, either. 8(
Uwe
Disable ALL shaders, disable hires textures in launcher, set res to 1024x768. You’ll get 60 Hz in 3d rather than 25.
Edit: also hide the window, see edited first post. Also disable autogen for trees (slider in graphics menu).
Edit: my earlier statement
You can run both on 3 CPU affinity threads with little overlap. That’ll offload some processing.
Is incorrect. Only run the client with 3 affine threads. The server must use all cores for the pseudo-rendering job.
Edit: Please investigate for me absurdly high FOV. It should lower LOD detail used.
-
I cannot set the resolution in the graphics section to anything other than the native rdp window resolution. I checked the gfx settings and they looked fine (I found some additional shaders still enabled, but hires textures and most of the other stuff was off).
Any idea how to hard-code the BMS resolution used?
All the best, Uwe
-
Edit: my earlier statement Is incorrect. Only run the client with 3 affine threads. The server must use all cores for the pseudo-rendering job.
Edit: Please investigate for me absurdly high FOV. It should lower LOD detail used.
i can’t start server and client concurrently. i tried to clone BMS root folder, but without success. BMS client warns me it can’t start because another instance of itself is running.
-
i can’t start server and client concurrently. i tried to clone BMS root folder, but without success. BMS client warns me it can’t start because another instance of itself is running.
Hm, you don’t need to start a separate server when hosting on your gaming machine (if that’s what you are trying to do).
Uwe
-
exactly. i want to see if some performance gain can be obtained with a such method.
imagine a very large TE or even a campagin . -
I guess you’ll be limited by upstream b/w with many online pilots before you’ll run into cpu bottlenecks on a gaming machine.
Naturally, you’ll see better performance with only one instance of BMS running on a machine than you’d observe with two.
Cheers, Uwe
-
It can be usable for localhost. 4-core sandy bridge has 8 threads which only requires 3 for client. Server can use 4.
I assume BMS uses a hardcoded mutex name for the one instance check. I see two solutions here:
- run server and client as two separate Windows users and fast switch between them
- deal with it the other way, which I can’t condone here
- kill it with fire using a program
For the mutex name look for session objects with Windows tools.
sh
-
Did anyone using this server experience problems wherein A-G kills only registered as ‘damaged’ in the debrief, after the campaign was active for a while?
Using this server we have had this ‘bug’ pop up consistently in two separate campaigns. BfB standalone and ITO. -
There’s experimental functionality for “mesa_glthread” that speeds up time taken for “rendering”. It doesn’t decrease CPU usage but will leave more time for logic like physics, unit behavior, campaign stuff. Anyone wants this?
Edit: for my usage - https://cgit.freedesktop.org/~mareko/mesa/commit/?h=glthread-5-rebased&id=970a000921a75c2f3c8e5736106e7eae7a625145
TODO: need enable background callability for Windows GDI. Right now it’s only enabled for Unix DRI.