MilkyWay@home

[gomera]

no milkyway ma pekne sklamal, co sa kreditu tyka:
1. co je to za hovadina, aby bol kredit obmedzeny zhora
2. preco moj stary P4 3.0GHz dostava na jedno jadro (ma HT) max. pocet bodov rovnako ako 4 jadrovy AMD 1.9GHz (samozrejme jedno jadro je rychlejsie ako P4 )

ten kredit uz nema vobec ziadnu vahu v tom projekte, ked pomaly vsetky kompy dostavaju rovnako body, navyse, ak s tou opt. aplikaciou sa dostava neskutocne mnozstvo bodov v porovnani s ostatnymi projektmi - by som ich vyhodil z BOINC statistik (a ked nie vyhodil, aspon by sa do BOINC Combined nemali ratat)

som to len zacal ratat, ale uz koncim, to je humus (navyse nevedia dat dokopy normalnu aplikaciu)

BTW ten projekt je v alpha verzii - pocitaju sa body z alpha a beta verzii normalne do statistik (pozeram, ze ano, ale ...) ?

[Kiwi]

Tak ako pises. Ci ti bezi jadro na 1 GHz alebo 4 GHz, dostavas fixnych 2593 bodov per core.

[matob]

Vyšiel nový film: MilkyWay - Honba za kreditom

[Kiwi]

No hodili by sa dva Core i7-ky, dokopy 16 jadier * 2593 = 41400 RAC.

[gomera]

mam pocit, ze sa netreba namahat s tolkymi jadrami, moznoze len staci urobit niekolko virtualnych masin a kazdemu povolit 4 jadra a dostanes mozno rovnako
v kazdom pripade system pridelovania kreditov je teraz uplne o nicom; vyvolava to len zle reakcie na projekt samotny

[matob]

November 18, 2008
I've just finished implementing a bunch of optimizations suggested by our users in the new milkyway application (we're hoping to roll it out on wednesday). On our AMD Linux cluster it's running about 2-3x faster. I'm currently getting some numbers for OSX and Dave is working on some numbers for windows. We should be releasing the code wednesday when we put the app on our server.
--Travis

[Honza]

Dulezita a nezodpovezena otazka je, zda-li je to 2-3x rychlejsi nez "optimalizovane" nebo nez optimalizovane (a tudiz stale radove pomalejsi). Optimalizovane davam do uvozovek, protoze i ty maji stale jeste rezervi na zrychlovani.
Jestli rychlejsi nez oficialni, tak mi to prijde po mesicich peripetiji ubohe. Jestli nez rychlejsi nez "optimalizovane", tak je to "uspech" s pachuti, jakou lide pamatu na SZTAKI. Protoze to co se pocatilo rok, se dalo spocitat za par dni.

[Kiwi]

Bolo by vhodne urobit nejaky optimalizacny checker, ktory "prejde" cely kod a vypoti, da sa to
este optimalizovat, alebo je to optimalizovane na maximum. Ale to je asi hudba buducnosti.

[Honza]

Hmm, a jako prvni bych to pustil na Windows...

[Kiwi]

Ja ked som niekedy robil kod, tak ak sa dal 16 kB kod vypoteny prekladacom a linkerom
upravit ma 512 alebo 1024 bajtovy, tak som nad tym sedel cele dni. Podobne ak islo
o casovu optimalizaciu. Dnes sa uz programuje tak, ze nad clovekom stoji otrokar
a dolezite je, aby sa to aspon tvarilo, ze to chodi bezchybne. Len dost dobre nechapem,
na kyho certa som sa ucil dokazy spravnosti algoritmov ?

[gabberattack]

Hmm, a jako prvni bych to pustil na Windows...

Windows ? To je co ?

[Palo M.]

Ja ked som niekedy robil kod, tak ak sa dal 16 kB kod vypoteny prekladacom a linkerom
upravit ma 512 alebo 1024 bajtovy, tak som nad tym sedel cele dni. Podobne ak islo
o casovu optimalizaciu. Dnes sa uz programuje tak, ze nad clovekom stoji otrokar
a dolezite je, aby sa to aspon tvarilo, ze to chodi bezchybne. Len dost dobre nechapem,
na kyho certa som sa ucil dokazy spravnosti algoritmov ?

Ono je to dnes s programovanim este omnoho horsie, ako si myslis. Bud rad, ze nevies vsetky detaily. Mne je z toho niekedy poriadne na vracanie...

A spravnost algoritmov je pravdepodobne len akademicka zalezitost - v praxi som sa s tym este nestretol a mam silne podozrenie, ze sa vobec nepouziva (co samotne nie je tragicke, lebo v skole sa clovek uci vela veci, ktore potom aj tak nepouziva).

[matob]

November 19, 2008
I just implemented another round of optimizations, and now on the AMD Linux cluster it's running 8x-9x faster (almost 4 times as fast as the previous version). The machine I'm testing on is really high end, so I'm hoping that older archiectures will get even more improvement from this. Also, we're still on schedule to put the new app on the server this afternoon and start with the new workunits.
--Travis

November 20, 2008
Nate came to us with some problems in the current application code, so we've been working on fixing that. Everything is looking right (or very close to it) at the moment, so we'll be trying to get the new application out this afternoon (around 2pm EST).
--Travis

November 20, 2008
We've put the new version of the application up on the server, currently we only have it for windows. I'm working on compiling the application for OSX at the moment. Afterwards I'm going to try to start a few workunits tonight and see how horribly things crash.
--Travis

[Kiwi]

Je tam aj zaujimava diskusia na fore ohladom dlzky WU, kreditu a "optimalizacie".

[matob]

November 21, 2008
We've flagged the new application as a test application, so you can opt out of testing it. Default preferences are to NOT run test applications, so if you want to help us test it you can enable this in your project preferences.

November 21, 2008
We've fixed the problem with the application and have recompiled it for windows and OSX. I'll be starting up some workunits tonight (in the next few hours) and hopefully they'll work ok. If the workunits don't fail, I'll start up the new assimilator.
--Travis