| |
| |
| |
|
Page: 1 2 3 4 5 6 7 8
Comments:
<0> hiya
<1> http://www.youtube.com/watch?v=NBV52GPhNjw
<2> hahahaha, that's far better then the .gif
<2> the guy did like, grab-hug him from behind, so the headbutt seems a lot more reasonable now
<3> remap_file_pages() is rather nice for efficient multiple mmap()'ing on Linux. Does anyone know something similar on FreeBSD?
<1> never used it
<1> tehse new syscalls are of so marginal use
<3> To me, it seems to be the efficient solution for a large and randomly accessed database
<1> what exactly you can do with remap_file_pages() that yuo can't do without ?
<3> Non-linear file to memory mapping
<1> but why it's better than linear ?
<1> i don't get it
<3> You can manage without it, but there are performance issues
<1> looks veeeeeeeeeeeeeeeeeeery marginal to me
<1> show me 1 app that wins using it
<1> and loses without it
<1> there are rather few apps that use mmap in the first place
<3> No idea yet, I just discovered it, but I plan to try it out with my database
<1> i dunno. looks to me i could do the same with several mmaps
<1> mapping a portion of the file with each mmap
<3> Hrm, the function prototype is fairly wrong on that page
<1> ok, i see. it saves some kernel memory, but only
<1> if you do several mmaps
<1> if you don't do several mmaps, but just use 1 mmap for the whole file, it does not save any memory
<3> Correct, but I can't quite rely on mmap()'ing a 20gb file
<1> hmm
<1> on 64-bit machine, you can
<4> aeon mish
<1> ay ay
<1> and if you do 20-gb files, you are sooooooooooo likely to have 64-bit machine
<1> yeah, mish was, to some degree, good nick, but the mis- prefix was bugging on me
<4> i have 20gb files, and i dont have a 64bit machine :P
<4> Misanth Mister_X
<3> It's all on 64 bits, although I'm not presently mmaping the whole database. I suppose that might be preferable even
<1> yeah, mis-aligned mis-demeanor
<1> <3 64 bits
<1> <3 dual-cpu
<4> not me. i prefer one really bad*** cpu.
<6> hello
<4> theres nothing two cpus offers that one cpu with twice the capacity doesn't offer.
<1> but where do you get teice the capacity cpu, when dual 3HMZ cpu isnot that expensive, but 6GHZ?
<4> if i hear one more person say 'dual core makes your computer multitask better' im going to go on a killing spree
<1> but where do you get 6MHZ cpu ?
<7> doesn't it? :)
<4> right, the nonexistance of n*2 hz processors is just a technicality
<1> AaronWL: "making multitask better" is totally NOT the point of dual-cpu
<7> what is the point of dualcpu
<4> multicore/multiprocessor is vastly inferior to single.. its just a hack because single don't go fast enough.
<1> AaronWL: i'm of totally opposite view
<1> i wait to the time when adding another bunch of cpus to your comp is like adding ram today
<4> dual core cpus are advertised to consumers primarily on superior multitasking ability.
<4> at least in the US
<4> which is complete nonsense
<1> i'm technical enough to buy things NOT based on what they are advertised
<4> Aeon_, problem is that it is extremely difficult (read: expensive) to parallelize many common algorithms.
<1> let's talk price/performance, OK ?
<3> I'm all for multiple cpus or cores... Personally, for my needs, I would even prefer 256 processors of 50mhz or so :)
<1> AaronWL: absolutely not in multi-client systems
<1> multi-client systems are so VASTLY easy to paralellize
<4> depending on the reason performance is needed, i prefer two computers to one computer with two cpus.
<8> Dual core is just a way of maximizing performance/price. It doesn't make sense to include in marketing for nerds.
<7> for a workstation, where a user is running 3-5 programs, multi-cpu has no advantage??
<7> I think most people realise that multicore won't speed up one serial program
<1> AaronWL: programmer always have a choise of multithreading vs multi-forking
<9> Maloeran: get 1024*768 cpus
<10> some resources, like memory, are shared between multiple CPUs
<8> Kardos: You don't know many "most people", then.
<1> if programmer chooses multi-forking, there's little difference b/w dualcore and two computers
<6> mmmm, multi-forking
<7> hmmm, could be
<4> the problem with throwing a lot of cpus into a computer is that all of the other components also are not being scaled up. eg, various busses, disk drives, network card, etc
<3> The problem with the whole x86/amd64 architecture is how memory synchronisation between processors is handled at the hardware level
<3> So you won't see 256 amd64 processors sharing a single memory bank, like Itaniums do
<6> Aeon_: i just ate so i'm god; i'l just play with this silverware for a while
<1> Numa!
<6> good*
<4> i guess the main advantage of a huge number of cpus is you get to save $$$ on RAM modules in the case where each thread needs the entire dataset.
<1> Maloeran: already 256 Itanium were produced & sold ? Wow
<4> but thats only provided that the entire dataset fits into RAM
<1> s/Itanium/Itaniums/
<4> im a huge fan of horizonal scaling of commodity hardware.
<1> AaronWL: you are huge fan of ... what ?
<11> as apposed to verticle scaling.
<11> vertical even
<8> Isn't verticle the opposite of horizonal?
<1> no
<12> howdy OT
<9> 1+1 = <3
<1> it's vertical, btw
<12> and voff.
<9> OrngeTide!
<11> if each thread takes 8M of stack and you have 1024 cpus. that's 8gb you need for stacks, although you would likely not use even a tenth of the total 8gb on average.
<11> hi zid
<1> 8M is huge
<11> zid_, i'm going to ship it for free. i think i figured out how to get my company to pay for it. :)
<11> Aeon_, it's the default on most systems.
<4> instead of having one machine with $100k hardware, have ten machines $1k hardware. horizonal scaling.
<1> you shan't have 8MB stack if you want lots of threads
<9> ahaha awesome
<11> the idea being you should never run out of stack
<9> it's a business deal man, one hdd and I add a newline to your .c files to stop the warnings, k? :P
<11> zid_, i dunno if there will be any duties or whatever your evil goverment does
<1> if you want lots of threads, why don't you decrease stack size to minimum
<9> imported my DS fine
<9> it was labeled "books" though
<9> :D
<11> zid_, you add #include <stdlib.h> so i don't have to cast malloc anymore
<11> zid_, ahah. nice
<9> I like casting malloc
<1> it's default but you *have* to decrease it if you want >256 threads
<4> or you just need a large address space
<1> because 256 threads and default stack size overflows 4GB virtuial mem size
<9> OrngeTide: Do you play Kingdom of Loathing?
<11> what should i label it? "Adult Books" ?
<4> i dont see any reason why each thread needs its entire stack space paged-in
<11> yea. i play it once in a while. i haven't signed on in over a two months though
<9> i recommed "preteen pornography" or "dead babies"
<9> 1 month = account go bewm unless you've earnt nodelete
<11> Aeon_, yea. i use 8kB stacks on my thread library. since i usually run around 6000 threads
<12> hm
<11> zid_, yea. i have nodelete
<12> should I bring my laptop for a 3 day vacation in london
<12> it'll only make my backpack heavier.
<11> zid_, "donor organs"
<9> OrngeTide: You should play some more man
<11> been too busy
<9> NO U AMN'T
<11> i'm writing some docs right now for my startup. kind of stumped on things
<12> OrngeTide: hey, do they have drug testing at cisco?
<9> You have a startup? Can I have 1 share?
<11> wow. i'm only level 10. i have 200 turns to spend. eheh
<12> zid_: at a nominal fee of $5000. yes.
<11> zid_, no. shares are $250k each
<12> OrngeTide: I sold one stock for $500, you should give me a bonus
<11> SgtUnix, uh. no. they are acually pretty expensive
<13> OrngeTide: you're doing a startup?
<11> Wazm, yea. why you need a job? :P
<9> OrngeTide: I do
<12> zid_: but he won't hire you. get lost.
<13> OrngeTide: depends, how much VC funding have you secured? :-)
<11> zid_, i can only hire people who work. sorry.
<9> I work god damnit
<11> Wazm, nearly a million dollars. by nearly i mean zero.
Return to
#c
or
Go to some related
logs:
#mirc
#hardware
ordinal nvapi.dll
steadyshells
jmandon
chrome begin end
#politics
dog butholes
#visualbasic
#beginner
|
|