shrike 0 Posted November 12, 2002 For tuning purposes I wanted to put some load on our server without connecting it to the internet. We just connected with one user over LAN and used Coop Missions with lots of AI, forcing some action by sending AI units towards the enemy positions. Ist this method any good to create a server load comparable to a few "real" players being connected? With roughly 48 AI on the map + 4 or 5 manned tanks, all shooting and shelling like crazy the server FPS (#monitor) were around 48-50. Is this good? Is this bad? I have no clue, it's my first dedicated server. It's running on Linux, OFP version 1.88b. Please enlighten me... Thx Shrike Share this post Link to post Share on other sites
suma 8 Posted November 12, 2002 </span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE">We just connected with one user over LAN and used Coop Missions with lots of AI, forcing some action by sending AI units towards the enemy positions. Ist this method any good to create a server load comparable to a few "real" players being connected? <span id='postcolor'> No, it is not comparable. A lot of server CPU load is not caused by units simulation (which is what you are doing), but with traffic control (determining what should be send to which player). </span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE">FPS (#monitor) were around 48-50. Is this good? Is this bad?<span id='postcolor'> This is the best fps possible, as server fps is limited to 50. When running at 50 fps, you should also watch CPU load by using linux command "top". Fps 15 and above is usually considered playable, 25 and above is very good. Share this post Link to post Share on other sites
shrike 0 Posted November 12, 2002 </span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE"> No, it is not comparable. A lot of server CPU load is not caused by units simulation (which is what you are doing), but with traffic control (determining what should be send to which player). <span id='postcolor'> I guessed so...I will have to wait for a stress test until it's online then... </span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote </td></tr><tr><td id="QUOTE"> This is the best fps possible, as server fps is limited to 50. When running at 50 fps, you should also watch CPU load by using linux command "top". Fps 15 and above is usually considered playable, 25 and above is very good. <span id='postcolor'> I knew it tops out at 50 FPS, I meant to ask if it was good in relation to the number of AI...which is irrelevant as you mentioned above. What I found rather surprising is that "top" gave me a pretty balanced CPU load (User) of around 15% to 35% for each of the 2 CPUs: As far as I remember OFP does not take advantage of MP systems! How come? There were only  peaks alternating between the CPUs, sometimes CPU 0 had a peak of around 80% sometimes CPU 1... System CPU load was around 0.2% on one of the both CPUs, the other was 0.0% I hope I am not asking dumb questions...Linux and Dual CPU systems are unexplored fields for me... Other than that the server worked great. No crashes, no squad.XML probs, nothing.... Thanks a lot anyway for your quick reply! Share this post Link to post Share on other sites
suma 8 Posted November 12, 2002 </span><table border="0" align="center" width="95%" cellpadding="3" cellspacing="1"><tr><td>Quote (Shrike @ Nov. 12 2002,12:34)</td></tr><tr><td id="QUOTE">What I found rather surprising is that "top" gave me a pretty balanced CPU load (User) of around 15% to 35% for each of the 2 CPUs: As far as I remember OFP does not take advantage of MP systems! How come? There were only  peaks alternating between the CPUs, sometimes CPU 0 had a peak of around 80% sometimes CPU 1... System CPU load was around 0.2% on one of the both CPUs, the other was 0.0%<span id='postcolor'> This is probably due to fact that the Linux thread manager does not always assing the same CPU to the running process. On schedulling timeslice different CPU can (but need not) be assigned. It is very likely the total for both CPU will always stay under 50 % or very close, due to the fact OFP does not take advantage of MP systems. Share this post Link to post Share on other sites
whisperFFW06 0 Posted November 12, 2002 Hi Suma, Could it be possible that this MP behaviour could explain the freezes observed on Linux servers w/ multiprocessors? (here, bottom of page) Something like lock pbs on shared memory.... Whis' Share this post Link to post Share on other sites
