| |
| |
| |
|
Page: 1 2 3 4 5 6 7 8 9 10
Comments:
<0> no, but when chuck norris swears i kick his ***
<0> :/
<1> Kid: no, he's trying to mock me
<0> nah
<1> good boy
<2> fighting a bot or playing with it?
<1> not that it's trivial to make 5k-clients-serving server with poll()
<3> reason for performance benefit is epoll is O(1) while poll is O(n) for going through the fd's. from what i read
<1> flexing
<3> epoll uses kernel notification
<3> noidea about the original question though
<3> :)
<2> .me sprays rip with troll remover
<1> if i really hit the poll() performance wall, i'd write my owen kernel mechanism similar to win32 completionports
<2> oops :)
<1> but i didn';t , so far
<4> raymond|: how to expand the original fd created by epoll_create()
<4> i.e. I did epoll_create(10) and I want to watch for 5 more
<1> pha: -> #kernel
<1> pha: most probably, you need to delete this epoll and create new, bigger, one
<1> check with kernel sources
<4> but then what happened to all the added ones?
<1> who cares
<1> nobody's using this api anyway
<4> :\ lost clients?
<4> :P
<1> it's dead child
<4> you sure?
<1> yes
<1> maybe not even the guy who wrote this kernel piece
<4> why are there native man pages on it then?
<1> my system has ~5000 apis manpages, doesn't mean i must use all of them
<1> if you want to use it, use it on your on trouble
<1> don't complani to *me*
<1> i don't force you to use it, nobody forces you to use
<1> if you want to use it, handle it yourself
<1> you *can* bitch, but don't expect condolences
<1> expect "you deserve it" grin
<4> ;)
<0> heh
<0> :)
<0> is there any point to using judy arrays at all if the hashes it will be replacing are all less than 256 buckets?
<3> http://www.xmailserver.org/linux-patches/epoll_create.txt
<1> wtf is judy arrays ?
<3> says it's just a hint for the size of the table
<3> so you should be able to go over it?
<0> its a cache optimized trie, ***ociative array
<3> and it'll resize, maybe double when you overshoot, <-- guess
<0> http://judy.sourceforge.net/
<0> this is too early in the morning to actually think.
<1> i like incremental hshtables
<0> this is supposed to be the fastest ***ociative array around from what i gather
<1> they avoid rehash the whole hashtable when table grows, and adds ~5 lines of code
<1> i still think hash is the fastest
<0> benching says otherwise afaik
<0> but thats only with hashes over 256 buckets
<0> im about to do my own for less than 256
<1> i gotta print & read em
<0> text terminal?:/
<1> no, paper
<1> will read in the toilet :-)
<0> hehe
<0> good idea, i get my best ideas while in there doing either ****ting or shaving or showering
<0> :/
<5> yeah
<5> me too
<1> rip: ok, having read the description of judy arrays, i see
<1> that it claims it's faster than *trees* (avl trees, b-trees)
<1> nowhere is claims it's faster than hashes
<1> hashes being O(1), and b-trees and judy arrays being O(log N)
<1> my conclusion isjudy is not competitor to hashes, and it's nowhjere faster than hashes
<0> en.wikipedia.org/wiki/Judy_array
<0> and http://www.nothings.org/computer/judy/ has some benches
<0> (nobody ever saud it was faster always either tho)
<6> Mornings!
<6> even though it's 2pm
<0> its 9am here and i havent slept yet
<0> im so cool.
<6> diddums
<0> :/
<0> out last night, since youre so late up?
<6> no
<6> althought acctualy I was *outside* last night
<6> but I don't think that counts as going out
<0> hehe
<0> well, i was out clubbing on friday since its formula 1 grand prix weekend here, and it ended with me waking up 8PM saturday, which is why im still up :P
<0> Mish_: from doc/index.html at judy.sf.net : However, JudyHS will perform better than a hashing method with smaller and larger populations than the optimum hash table size. JudyHS does not have a degenerate performance case where knowledge of the hash algorithm can be exploited. (I.E. JudyHS does not use a linked list to handle hash collisions, it uses a tree of JudyL arrays and a virtual hash table size of 4 billion).
<0> the obvious advantage is that it doesnt need tuning like a hash table
<0> oh and the same page also says: Judy arrays are both speed- and memory-efficient, with no tuning or configuration required, across a wide range of index set types (sequential, periodic, clustered, random). Judy's speed and memory usage are typically better than other data storage models such as skiplists, linked lists, binary, ternary, b-trees, or even hashing and improves with very large data sets.
<7> And Judy won't lie about Judy! they're not biased!
<0> that's another issue all in all which is part of why i asked here in the first place:p
<0> but he concluded that the documentation didnt say anything about being faster than hashes
<0> i was merely proving the oposite
<3> i'm sick of the word judy already
<1> well this judy thing is abviously O(log N) how can it be faster than hashes ?
<1> *obviously
<7> It's O(0) of course
<7> Yes, 0
<3> the tree nodes are identified by a byte(s) taken from the key, i think it works more like a hashtable but i got too sick of reading it to continue
<3> i think it's O(1) with reduced constant by improving cache misses or something
<3> hits rather
<3> it works like a regular hashtable until 256 entries or so, where it factors out the
<3> MSB to identify a node in the tree or something weird
<3> i don't know really, i'm losing the will to live
<3> am i going crazy or is figure B-1. The caching curve a bit misleading: http://docs.hp.com/en/B6841-90001/apb.html
<3> i see a picture of 2 kids, maybe i need some sleep
<8> :D
<8> someone hacked the hp server...
<3> ok, you see them too :)
<8> :)
<9> is it possible to have a negative in a unsigned char?
<10> it's unsigned
<10> it can only represent positive numbers
<9> when i set its value to -5 it shows as 251, how does 251 represent -5?
<10> this is an unsigned char then
<9> yeah
<10> what range of numbers can a char hold?
<7> pha: Try /cycle
<9> 0-255
<10> 0, -1 (255), -2 (254), -3 (253), -4 (252), -5 (251)
<10> see a pattern here?
<9> oh i see
<9> well this is for FLAGS emulation in a cpu core, so if i check if the number is >=127 then its actualy a negative
<3> 127 is +ve
<3> >127 is -ve
<9> oh >127 then oops
<3> in that it wraps around
<7> The N flag is the value of the high bit. You can use AND to get that value
<7> Alternatively, shift will also provide it
<7> (Z80 or C64, paulz?)
<7> Er, 6502, rather. :/
<11> raymond|: In some parts of the world, showing knees is considered offensive. :-)
<9> z80
<7> Yeah, you can just set N to (A&128)?1:0
<1> Teckla, who offensed you ?
<10> gameboy is a z80
<9> the >127 is the check for overflow too right?
<1> *offended
<7> The gameboy is just about a z80
<10> and uh ... sega master system
<10> yeah it'
<10> s not a true z80
<7> paulz: The O flag is.. a little more complicated
<10> WE KNOW MR.SMARTYPANTS
<10> oh and Colecovision
Return to
#c
or
Go to some related
logs:
#gentoo
exctascy
#linuxhelp
#stocks
software raid 1 slackware 10.2
#nhl
#unix
#stocks
#beginner
#nhl
|
|