{"id":103157,"date":"2022-02-01T15:36:44","date_gmt":"2022-02-01T13:36:46","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/metod-identifikaczii-sistemy-polzovatelya-na-osnove-informaczii-o-gpu"},"modified":"2022-02-01T15:36:44","modified_gmt":"2022-02-01T13:36:46","slug":"metod-identifikaczii-sistemy-polzovatelya-na-osnove-informaczii-o-gpu","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/metod-identifikaczii-sistemy-polzovatelya-na-osnove-informaczii-o-gpu","title":{"rendered":"Metoda identifikace u\u017eivatelsk\u00e9ho syst\u00e9mu na z\u00e1klad\u011b informac\u00ed o GPU","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
V\u00fdzkumn\u00edci z Ben Gurion University (Izrael), Lille University (Francie) a University of Adelaide (Austr\u00e1lie) vyvinuli novou techniku identifikace u\u017eivatelsk\u00fdch za\u0159\u00edzen\u00ed prost\u0159ednictv\u00edm m\u011b\u0159en\u00ed parametr\u016f v\u00fdkonu GPU v webov\u00e9m prohl\u00ed\u017ee\u010di. Metoda nazvan\u00e1 \u201eDrawn Apart\u201c se zakl\u00e1d\u00e1 na vyu\u017eit\u00ed WebGL k z\u00edsk\u00e1n\u00ed profilu v\u00fdkonu GPU, co\u017e v\u00fdrazn\u011b zvy\u0161uje p\u0159esnost pasivn\u00edch metod sledov\u00e1n\u00ed, kter\u00e9 funguj\u00ed bez pou\u017eit\u00ed cookies a bez uchov\u00e1v\u00e1n\u00ed identifik\u00e1tor\u016f v syst\u00e9mu u\u017eivatele. <\/p>\n
Metody zohled\u0148uj\u00edc\u00ed p\u0159i identifikaci charakteristiky vykreslov\u00e1n\u00ed, GPU, grafick\u00e9ho z\u00e1sobn\u00edku a ovlada\u010d\u016f byly pou\u017e\u00edv\u00e1ny ji\u017e d\u0159\u00edve, ale byly omezeny na rozli\u0161ov\u00e1n\u00ed za\u0159\u00edzen\u00ed pouze na \u00farovni r\u016fzn\u00fdch model\u016f grafick\u00fdch karet a GPU, tedy mohly b\u00fdt pou\u017eity pouze jako dopl\u0148kov\u00fd faktor pro zv\u00fd\u0161en\u00ed pravd\u011bpodobnosti identifikace. Kl\u00ed\u010dov\u00fdm rysem nov\u00e9 metody \u201eDrawn Apart\u201c je, \u017ee se neomezuje na rozli\u0161ov\u00e1n\u00ed r\u016fzn\u00fdch model\u016f GPU, ale sna\u017e\u00ed se odhalit rozd\u00edly mezi identick\u00fdmi GPU stejn\u00e9ho modelu, kter\u00e9 jsou zp\u016fsobeny heterogenitou v\u00fdrobn\u00edho procesu \u010dip\u016f ur\u010den\u00fdch pro masivn\u00ed paraleln\u00ed v\u00fdpo\u010dty. Bylo uvedeno, \u017ee variace vznikaj\u00edc\u00ed b\u011bhem v\u00fdroby umo\u017e\u0148uj\u00ed vytv\u00e1\u0159et jedine\u010dn\u00e9 otisky pro stejn\u00e9 modely za\u0159\u00edzen\u00ed.
<\/center> <\/p>\n
Uk\u00e1zalo se, \u017ee tato rozli\u0161ov\u00e1n\u00ed lze odhalit prost\u0159ednictv\u00edm po\u010d\u00edt\u00e1n\u00ed blok\u016f vykon\u00e1v\u00e1n\u00ed a anal\u00fdzy jejich v\u00fdkonu v GPU. Jako primitiva k identifikaci r\u016fzn\u00fdch model\u016f GPU byly pou\u017eity kontroly zalo\u017een\u00e9 na sad\u011b trigonometrick\u00fdch funkc\u00ed, logick\u00fdch operac\u00ed a v\u00fdpo\u010dt\u016f s plovouc\u00ed \u010d\u00e1rkou. Pro identifikaci rozd\u00edl\u016f ve stejn\u00fdch GPU se hodnotil po\u010det soub\u011b\u017en\u011b vykon\u00e1van\u00fdch vl\u00e1ken p\u0159i vykon\u00e1v\u00e1n\u00ed vrcholov\u00fdch shader\u016f. P\u0159edpokl\u00e1d\u00e1 se, \u017ee zji\u0161t\u011bn\u00fd efekt je zp\u016fsoben rozd\u00edly v teplotn\u00edch re\u017eimech a spot\u0159eb\u011b energie r\u016fzn\u00fdch exempl\u00e1\u0159\u016f \u010dip\u016f (d\u0159\u00edve byl podobn\u00fd efekt prok\u00e1z\u00e1n pro CPU \u2014 identick\u00e9 procesory p\u0159i vykon\u00e1v\u00e1n\u00ed stejn\u00e9ho k\u00f3du vykazovaly r\u016fznou spot\u0159ebu energie). <\/p>\n
Vzhledem k tomu, \u017ee operace p\u0159es WebGL se prov\u00e1d\u011bj\u00ed asynchronn\u011b, nelze pro m\u011b\u0159en\u00ed doby jejich vykon\u00e1n\u00ed p\u0159\u00edmo pou\u017e\u00edt JavaScript API performance.now(). Proto byly navr\u017eeny t\u0159i triky pro m\u011b\u0159en\u00ed \u010dasu: <\/p>\n
\n
onscreen \u2014 vykreslov\u00e1n\u00ed sc\u00e9ny v HTML canvas s m\u011b\u0159en\u00edm \u010dasu vyvol\u00e1n\u00ed callback funkce, kter\u00e1 byla nastavena prost\u0159ednictv\u00edm API Window.requestAnimationFrame a je vyvol\u00e1na po dokon\u010den\u00ed vykreslov\u00e1n\u00ed.\n
offscreen \u2014 pou\u017eit\u00ed workeru a vykreslov\u00e1n\u00ed sc\u00e9ny do objektu OffscreenCanvas s m\u011b\u0159en\u00edm doby prov\u00e1d\u011bn\u00ed p\u0159\u00edkazu convertToBlob.\n
GPU \u2014 vykreslov\u00e1n\u00ed do objektu OffscreenCanvas, ale pou\u017eit\u00ed pro m\u011b\u0159en\u00ed doby s vyu\u017eit\u00edm WebGL \u010dasova\u010de, kter\u00fd zohled\u0148uje d\u00e9lku prov\u00e1d\u011bn\u00ed sady p\u0159\u00edkaz\u016f na stran\u011b GPU. <\/ul>\n
P\u0159i vytv\u00e1\u0159en\u00ed identifik\u00e1toru na ka\u017ed\u00e9m za\u0159\u00edzen\u00ed se prov\u00e1d\u00ed 50 kontrol, p\u0159i\u010dem\u017e ka\u017ed\u00e1 zahrnuje 176 m\u011b\u0159en\u00ed 16 r\u016fzn\u00fdch charakteristik. Experiment, b\u011bhem n\u011bho\u017e byly shrom\u00e1\u017ed\u011bny informace o 2500 za\u0159\u00edzen\u00edch s 1605 r\u016fzn\u00fdmi GPU, uk\u00e1zal zlep\u0161en\u00ed \u00fa\u010dinnosti kombinovan\u00fdch metod identifikace o 67 % p\u0159i p\u0159id\u00e1n\u00ed podpory Drawn Apart. Konkr\u00e9tn\u011b pr\u016fm\u011brn\u00e1 doba identifikace kombinovan\u00e9 metody FP-STALKER byla 17.5 dn\u00ed, zat\u00edmco v kombinaci s Drawn Apart se doba identifikace zv\u00fd\u0161ila na 28 dn\u00ed.
<\/center> <\/p>\n
\n
P\u0159esnost rozd\u011blen\u00ed 10 syst\u00e9m\u016f s \u010dipy Intel i5-3470 (GEN 3 Ivy Bridge) a GPU Intel HD Graphics 2500 v testu onscreen byla 93 %, zat\u00edmco v offscreen \u2014 36,3 %.\n
Pro 10 syst\u00e9m\u016f Intel i5-10500 (GEN 10 Comet Lake) s grafickou kartou NVIDIA GTX1650 byla p\u0159esnost 70 % a 95.8 %.\n
Pro 15 syst\u00e9m\u016f Intel i5-8500 (GEN 8 Coffee Lake) s GPU Intel UHD Graphics 630 \u2014 42 % a 55 %.\n
Pro 23 syst\u00e9m\u016f Intel i5-4590 (GEN 4 Haswell) s GPU Intel HD Graphics 4600 \u2014 32,7 % a 63,7 %.\n
Pro \u0161est smartphon\u016f Samsung Galaxy S20\/S20 Ultra s GPU Mali-G77 MP11 v testu s v\u00fdstupem na obrazovku byla p\u0159esnost identifikace 92,7 %, zat\u00edmco pro smartphony Samsung Galaxy S9\/S9+ s Mali-G72 MP18 \u2014 54,3 %. <\/ul>\n
<\/center> <\/p>\n
Zaznamen\u00e1v\u00e1 se, \u017ee na p\u0159esnost m\u011bla vliv teplota GPU, a u n\u011bkter\u00fdch za\u0159\u00edzen\u00ed zp\u011btn\u00e9 restartov\u00e1n\u00ed syst\u00e9mu vedlo k deformaci identifik\u00e1toru. P\u0159i pou\u017eit\u00ed metody v kombinaci s dal\u0161\u00edmi metodami nep\u0159\u00edm\u00e9 identifikace m\u016f\u017ee b\u00fdt p\u0159esnost v\u00fdrazn\u011b zv\u00fd\u0161ena. Tak\u00e9 se pl\u00e1nuje zlep\u0161it p\u0159esnost vyu\u017eit\u00edm v\u00fdpo\u010detn\u00edch shader\u016f po stabilizaci nov\u00e9ho API WebGPU. <\/p>\n
Spole\u010dnosti Intel, ARM, Google, Khronos, Mozilla a Brave byly o probl\u00e9mu informov\u00e1ny ji\u017e v roce 2020, ale detaily metody byly zve\u0159ejn\u011bny a\u017e nyn\u00ed. Mezi jin\u00fdm zve\u0159ejnili v\u00fdzkumn\u00edci p\u0159\u00edklady napsan\u00e9 v JavaScriptu a GLSL, kter\u00e9 mohou fungovat s v\u00fdstupem i bez zobrazen\u00ed informac\u00ed na obrazovce. Tak\u00e9 byly zve\u0159ejn\u011bny datov\u00e9 sady pro syst\u00e9my s GPU Intel GEN 3\/4\/8\/10 pro klasifikaci extrahovan\u00fdch informac\u00ed v syst\u00e9mech strojov\u00e9ho u\u010den\u00ed. \n Zdroj: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"
![\"Metoda](\"\/wp-content\/uploads\/2022\/02\/947fc4477ff99c14ca7854e36e4ad976.jpg\")
<\/center> <\/p>\n![\"Metoda](\"\/wp-content\/uploads\/2022\/02\/c57096ca8d5833df92cd7f992f19a289.jpg\")
<\/center> <\/p>\n![\"Metoda](\"\/wp-content\/uploads\/2022\/02\/e5462f5b761dc6e16a43ac00d8f0b7aa.jpg\")
<\/center> <\/p>\n