Last night we were hosting a 30h testing session with 30 slots for Renault Clios at Mills Inner Loop C. At some point the pings started to get really bad, 500 ms and more and everyone was warping all over the place. Bandwidth wasn´t the problem, I was constantly monitoring it, and it never even came close to 10 Mbits/s (root server has 100 Mbit/s), up- and downstream were set 25 Mbits/s in the Multiplayer.ini. The server frame rate was fluctuating somewhere around 100 frames per second instead of 400. Looking at the CPU load showed that the server was attempting to use more than 50% CPU load. The server has an AMD Athlon II X4 CPU with four cores that clock at up to 2.3 GHz, so the server demanded more than 4.6 GHz for this single instance. That is quite a lot more than the 250 MHz suggested by ISI. The server population was never at the maximum of 30, it might have been slightly more than 20 at some point. I had to shut it down and start a new instance. With the new instance running I now monitored CPU load more closely and it seems to use quite a lot of CPU cycles for each client and it doesn´t seem to release those resources when clients leave. The highest population after the restart was 15 I think, so the CPU load wasn´t as high as before, ca. 40% this time. Coincidentally at some point many people left, with only three remaining but the CPU load stayed the same. Soon the server was completely empty but still using about the same amount of resources. Now the server is not a Windows server, it is running Debian GNU/Linux 6.0.6. a.k.a. Squeeze with kernel version 3.2.0 (amd64) from the backports repository. I am using WINE 1.4.1 to run rFactor2 Dedicated.exe. To be honest, I never expected native performance, a small(ish) overhead I had reckoned with, but not this. I believe it has nothing to do with WINE but I have no means to test on a native Windows host. I have a virtual machine using qemu-kvm on that Linux host, which runs Windows 7 on one virtual core. But that already uses about 20% CPU just when opening the dedicated server setup dialog as it does when running with WINE, so I guess I will see similar behaviour with that approach and wouldn´t even be able to host a 30 slot server. So, could those hosting from a native Windows host please have a look at your CPU usage graphs and share your experiences along with information about the host system? Also I´d like some feedback of those also using WINE, since I cannot be the only one according to this forum. And of course everyone is invited to join our server OcM-Racing to do some further testing. Thanks in advance and best regards, Marcus
We at SRVN.nl had a similair problem. Also running the Clio's but then on Malaysia North Loop with 28 clients. Everything went fine during the practice but when the server was set to qualification and also warmup the server FPS dropped to 40-60 instead of 400. When we started the race it went up to 400 and stayed that way during both races. Very strange. I believe it is happening since Build 125 but not on every circuit. LRP went perfect for example.
What does the CPU load look like in general, when there are say at least 10 clients connected? And what kind of hardware are you running? Edit: Are you running the dedicated server natively in Windows?
I didn't check the CPU-load, but whe are using Windows 7 Professional 32 . With a AMD Phenom x64 processor (3.4 GHz and 3.2 GB memory used)
Thanks for your reply. I would be really grateful if you could have a look at CPU load, when there is a considrable amount of clients connected (and driving), just so I can see if ISI were just overly optimistic with their statement of 250 MHz per server instance or if I was in thinking WINE would only introduce mild overhead.
It took a while but tonight we had the next round. In warmup and qualify the CPU load went up tot 30% of the system. That was with 30 clients and it was almost undriveable. When the server was set to race everything went normal again.
