| |
| |
| |
|
Page: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Comments:
<0> which we know it can be non-trivial to isolate the right place to fix such things
<0> i have seen people spend huge amounts of time optimizing something that was irrelevant, resulting in... no progress
<1> ok, I'll bite. just what is the performance problem you are having
<1> now granted. I'm not sure if I know a whole lot about webserver and tuning them
<2> you'd be better off using google than asking here
<3> LinuxQuest: well, if you keep 80% of the conversation inside your head where it belongs, that might help in future ;)
<4> schitzo: You don't want none of that
<1> or that kind of stuff
<0> slow service of web requests
<3> digital712: EXACTLY
<1> ok, a) define slow
<0> but in a funny way: on serving a large file, like an mp3, it doesn't pump it to the full bandwidth that it should
<5> safemode is there a difference between dual core and dual sockets? is dual core what is used for hyperthreading?
<6> no
<7> :-O
<1> what's the webserver doing? thread or process model? what's apache reporting it is doing when the reuqests are slow
<5> safemode i see. thanks
<0> it's one thing to pause before serving a small file; another is to serve the file, but serve it at low bandwidth
<0> it's forked, i believe php only wants to work in forked model
<6> hyperthreading is multiple execution cores, utilizing the same registers and memory bus and all that ****. dual core is exactly like multiple sockets, but is faster and more efficient
<7> Dual core is ... two CPU's inside one physical CPU (to simplify). Dual socket is two CPU's mounted on one motherboard. Hyperthreading is unrelated.
<0> hmm, i haven't even started looking at what apache is doing, and i will not trouble this chat with questions on that
<7> Well, not unrelated, but, er, I'll leave it to safemode. )
<2> did anyonee see the new demo's of Novel linux, sweet.....
<0> my first steps were to ask some basic-level questions: is the cpu a bottleneck? is the disk a bottleneck? is the ethernet a bottleneck?
<6> they got nifty pictures and **** online in various hardware places describing it
<2> love that eye candy
<0> i want to know those things before i start even digging in to apache, or mysql or php
<1> it's not a new linux. It's just Xgl stuff
<7> Yeah, try intel.com or amd.com, they have neat diagrams and stuff
<5> CeruleanD i see. so like i have two xeons, and they hyperthread, but that does not mean they are dual core?
<7> Right
<0> hyperthread is not the same as dual core
<1> LinuxQuest: I will almost bet you it's an application config somewhere and not hardware
<7> Dual core is relatively new (to the m*** consumer market)
<5> CeruleanD thank you very much. i've been confused about that since i bought it
<0> yeah i'm leaning that way also, but i would like to get some evidence if at all possible
<8> hyperthread is almost dual core. At least the OS thinks it's a dual core.
<2> Dual core is sweet for multitasking...
<0> i mean, it's even possible that there's a defective hardware component that is causing this. i doubt it but i would like to rule out base-level stuff before i spend a big chunk of time looking at smoething irrelevant
<7> Hyperthreading is just a marketing word for "lots of multi-threading"
<2> even on *shudder* windows
<0> HT appears as two cpus, and, if the compiler does just the right thing, it will allow two threads to almost sort of get the performance of having two cpus instead of one
<0> but it REALLY relies on some heavy compiler tricks
<0> i would guess that intel's compiler might be the best / only way to get the full advantage of HT
<0> alhtough i'm sure it's an active area of dev in gcc
<6> HT really doesn't require anything but multi threads to get any advantages. It's the schedular that may be tweaked to take advantage of HT over regular smp
<0> mm, not quite
<9> hi all
<6> yes quite
<1> OM!
<0> there are DEFINITELY some instructions that will cause one of the two threads running in the cpu to block
<9> hi schitzo
<5> LinuxQuest for certain apps i see definite useage of all four "virtual" cpus. two real, four hyperthreaded. like glx gears. i can run 4 instances of glx gears and a resource monitor like gkrellm appears that all four virtual cpus are working hard
<0> it has two pipelines
<5> and of coursetheres the responce time inder load. thats what i love.
<3> heya OM
<0> right, some apps can, but some types of mem access (or something) will cause one of the virtual cpus to dump its pipeline
<5> under
<5> LinuxQuest i see
<0> anyway max, there are some interesting articles about ht and performance. it's complicated stuff
<1> Cynic: are you bored?
<3> OldMonk: re- donut, i just get deputised occasionally ;)
<3> schitzo: very.
<5> LinuxQuest yes ive been googling for a few months now
<0> ideally it will get you almost the performance of two cpus, but in many real-world cases it does not, and in some real-world cases it performs WORSE than a single cpu
<1> hmmmm
<3> schitzo: got your tat yet?
<0> and it all depends a lot on compilers and the OS (scheduler)
<3> my friend ross has gone missing, and the cops are after him apparently, so... yeah. dunno when i'll get those images scanned for you.
<1> Cynic: no, I need to find someone to draw what I want
<5> LinuxQuest yes, thats what a server has to face , but as a desktop machine i find it rocks. i get desktop performance woithot any pre emptive patches so without hurting my throughput :)
<3> hmm.. well i know a good artist.. the web designer i work with is good, but i dunno if he does freebies like that
<1> Cynic: ahhh, not fun when the cops are looking for you.
<5> the tyan tiger is primarily a desktop board
<6> HT doesn't need special compiler directives. You can work around problems ***ociated with HT's short comings if you want. But most issues that will cause problems with HT are not things that can be worked around with ***embly, they're just functions of the program that are due to how it's programmed. Your process schedular can be tuned to keep threads that share certain types of memory regions bound to the same cpu rather than tr
<6> y to migrate them since they will block...etc. In any case, HT wil almost always be significantly slower than dualcore/smp systems.
<0> here's an article that talks about HT hurting per problems: http://news.zdnet.co.uk/0,39020330,39237341,00.htm
<0> safemode: yes
<0> it doesn't need compiler directives but it does need work to occur in the right order, i thikn order of mem access may be the big factor
<5> LinuxQuest thanks
<0> and cpu caching, etc. this is all weird stuff that i don't know much about
<0> maxheadrm: no prob. it's an interesting article
<0> oh and: http://news.zdnet.co.uk/hardware/chips/0,39020354,39198583,00.htm
<6> you can play with compiler **** all you want to fix P4's architecture problems.. but it's not something that needs to be done. Just sit there and face the fact that if you have a P4 based processor, you have a ****ty cpu
<0> i mean, HT is a great idea but it's not perfect
<5> cpu cacheing seems to be a big problem but all i can do about it is try and avoid it by turning off things that rely on it. im not a kernel programer
<0> well, max, what kind of ill effects are you observing? or is this just, "how fast can i get this thing to go?"
<1> HT ****s. we get better performance when we don't HT
<5> LinuxQuest what i love is how i can compile something, watch a video, and still have no loss of performance. without pre-emptive patches. it always has a processor to spare when i move my mouse or press a key.
<5> no need for tricks that hurt throughput
<6> get a cpu that doesn't have a 91021 step pipeline that ends up falling on it's face whenever anything isn't perfectly tuned for it.
<0> maxheadrm: simply, if it makes your desktop experience better, than it's good in your desktop computer
<6> and any preemptive low latency compiled kernel shouldn't have a problem compiling/ watching movies and not really losing responsiveness even on syngle cpu systems
<5> LinuxQuest only problem is my proprietary nvidia drivers lock up seldomly and i get a message that says : Failed to allocate mem resource #6:20000@f0000000 for 0000:01:00.0 on boot.
<0> ouch
<5> which is my agp video slot
<0> well, my desktop is an out-of-date single cpu thing
<0> my next will be a dual of some kind
<5> it hasnt happenned in a few weeks so it is possibly been solved
<5> LinuxQuest my old laptop bliterally burned out the motherboard and i figured i should spend a simmilliar amount on a desktop
<5> it might have been allergic to cola
<4> Man, I really need a utility to scan all IP addresses in my 192.168.0.100-199 range
<10> scan how?
<4> Mainly telnet port 22
<9> use nmap
<0> yeah, unfortunately i'm unlikely to spill anything on my desktop so i guess i'm stuck with it until i can't stand it
<3> reality_: nmap no good?
<0> reality_: nmap
<9> ouch!
<3> heh
<3> i think the consensus is nmap.
<11> hrm sweet it seems the OpenSwan 2.4.4 tarball can build functional debian packages finally.. it just calls them 2.4.0 which is easily fixable
<0> or you can just use tcpdump and see who is talking on the segment
<10> **** antec. incompetent bags of ****
<4> God, I hate installing nmap on windows
<4> Ah...well...looks like things have improved for the windows versoin
<3> s/nmap on//
<4> version
<0> i've never tried it on windows
<4> Cynic: Well, my linux box is sitting about 7 feet away from where the monitor cable reaches (monitors are mounted to the wall)
<4> Cynic: And I'm trying to get it's IP address to ssh to it
<3> ah, heh
<4> And I'm trying to do this without having to move either of the boxes around...the linux box is headless and keyboard less.
<3> i can't remember who it was, but a year or so ago someone was complaining that they had a box, that was up, and pinging, but they just couldn't figure out where in the apartment it actually was :P
<4> Cynic: bash.org
<3> yeah, probably got quoted
<12> i'
<4> Probably quoted from bash.org seeing as it's been up there for at least a good two years
<3> oh? well if it's not a #linuxer then i guess it was just being quoted from there.. that's slightly disappointing
<9> tsokolat!
<13> OldMonk!
<13> g'evening all
<3> evenin tsokolat
<14> hey tsokolat
<13> hiya..
<13> so quiet in here
<15> SWEET
<15> The bananaphone song is on my game right now!
<15> ring ring ring ring banana phone!
<10> ding dong ding dong ding dong ding danana phone?
<3> Steakk: i tried to get you a referral bonus but it seems it'll only give you L$2000 if i upgrade to premium
<16> poof
<0> ok, good night. thank you to those who actually helped me figure out what was going on, and to those who called me an idiot, well, people who lack confidence in themselves do that kind of thing
<15> Cynic: that blows
<3> Steakk: my name is Shep Spoonhammer if you see me around
Return to
#linux
or
Go to some related
logs:
yokmulan
#php
#MissKitten
naked buttoks
xarabank
#chatzone
#linux
#AllNiteCafe
Aceyalone feat. Riddlore - Find Out
#linux
|
|