-- [ Picked text/plain from multipart/alternative ] You're kind of there I think, not sure about the bit where you say 50% of one physical cpu though (which is typically the confusing bit, as you "can" set it to use 50% in some setups, just people rarely ever do as its not default).
Essentially one single thread "could" eat up 100% of a physical cpu and there would be no spare ht usability, so 1 virtual 100% cpu then unusable which is the problem (even though it can show 100% free, it doesn't know it can't use it). ht only has gains in certain situations, think on the list when someone managed to do a test with prime95 it was about 10% advantage with ht (i.e running 2 prime95 instances and timing them). It's a tricky one as well as I believe there's a couple of different top versions which show the cpus slightly different in overall usage, and may calculate it better, will try and get a test done at some point to see if its still the same. For me thats why I tend to use the load calculations more than cpu (can also highlight other issues), as its very hard to get a clear picture on cpu usage with ht enabled boxes. At least with load you have an idea how many processes are waiting for time etc. Everyone seems to have their preferred way of checking though :). On 11/15/06, k-mystik <[EMAIL PROTECTED]> wrote: > > I totally agree, the OS has no idea 2 of its CPUs are virtual. > > If I launch a single-thread process, it will eat up 100% of one logical > CPU (50% of one physical CPU), which is what it is supposed to do. > A multi-threaded process (4 threads for instance) will consume 100% of > both CPU; still ok. > > HT is just a way of cheating to be able to "process" 2 threads 50/50 on > the very same CPU. > > or maybe I'm totally wrong on this HT thing... > > > Ian mu wrote: > > -- > > [ Picked text/plain from multipart/alternative ] > > Your'e missing the point still a bit, with ht you don't have 100% global > cpu > > to play with in the first place, you only have 50% plus the benefit of > ht. > > The o.s essentially doesn't realise 2 of the cpu's are virtual and can't > be > > used to their full extent, it still thinks you have 4 full cpu's. This > bears > > our roughly with the load figures top right. Personally we try to keep > all > > of our dual cpu machines under a load of 2 (not one has gone over in the > > last week), but there's a fair bit of slack in that and can go higher, > just > > we prefer not to for performance reasons. > > > > I wouldn't ideally want to see double that (whilst things like > webservers > > would probably be ok with higher loads as less noticable) on that box, > let > > alone triple, but depends what kind of performance and contention people > are > > ok with really. > > > > > > > > On 11/15/06, k-mystik <[EMAIL PROTECTED]> wrote: > > > >> Right now, 49 people are playing on one of our machines, here's top > >> output: > >> > >> top - 22:09:17 up 43 days, 22:02, 1 user, load average: 1.82, 1.85, > 1.73 > >> Tasks: 97 total, 3 running, 94 sleeping, 0 stopped, 0 zombie > >> Cpu0 : 19.5% us, 1.7% sy, 0.0% ni, 73.5% id, 1.0% wa, 1.3% hi, > >> 3.0% si, 0.0% st > >> Cpu1 : 23.7% us, 1.7% sy, 0.0% ni, 74.2% id, 0.0% wa, 0.0% hi, > >> 0.3% si, 0.0% st > >> Cpu2 : 23.3% us, 2.3% sy, 0.0% ni, 74.1% id, 0.0% wa, 0.0% hi, > >> 0.3% si, 0.0% st > >> Cpu3 : 17.3% us, 1.3% sy, 0.0% ni, 81.0% id, 0.0% wa, 0.0% hi, > >> 0.3% si, 0.0% st > >> Mem: 2060228k total, 2028756k used, 31472k free, 145668k > buffers > >> Swap: 2031608k total, 88k used, 2031520k free, 1336232k cached > >> > >> Which is 20.95% global CPU average. > >> The max players we had at the same time was ~100 and the server had an > >> average CPU usage of 40% (spiking to 50%) > >> So, these numbers well reflects our calculations. > >> > >> -- > >> k-mystik! -- get2play administrator > >> web: www.get2play.com > >> irc: #get2play > >> > >> > >> Marcel wrote: > >> > >>>> Marcel wrote: > >>>> > >>>> > >>>>>> Hey, > >>>>>> > >>>>>> We use Fedora Core 4 on dual Xeon 3.2ghz (2Mb L2) and run without > any > >>>>>> problem 14 CS1.6 server (12 players) at stable 250fps. > >>>>>> (yes we recompiled the kernel) > >>>>>> > >>>>>> > >>>>>> > >>>>> And how many slots are filled max at which cpu usage? That's the > only > >>>>> thing that matters and I don't think all 168 Slots can be filled. > >>>>> > >>>>> > >>>> Based on our calculations, all 168 slots can be filled in with still > >>>> some CPU spare. > >>>> Currently a 10 slots CS1.6 @ 250fps uses up to 18% of one logical CPU > >>>> > >> (4 > >> > >>>> logical CPUs are present due to HT). > >>>> > >>>> Btw: if someone can explain the high context-switch (shown in "vmstat > >>>> 1") numbers I have (12000+): shoot! > >>>> > >>>> > >>> Have you tried that our or are these numbers only calculations? > >>> AFAIK the two HT-logical cpus can't handle the same "usage" as the > real > >>> physical cores. > >>> > >>> > >> > >> _______________________________________________ > >> To unsubscribe, edit your list preferences, or view the list archives, > >> please visit: > >> http://list.valvesoftware.com/mailman/listinfo/hlds_linux > >> > >> > > -- > > > > _______________________________________________ > > To unsubscribe, edit your list preferences, or view the list archives, > please visit: > > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > > > > -- > k-mystik! -- get2play administrator > web: www.get2play.com > irc: #get2play > > > > _______________________________________________ > To unsubscribe, edit your list preferences, or view the list archives, > please visit: > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > -- _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
