Ikhodi yenjini ye-BizingSQL SQL ivuliwe, kusetshenziswa i-GPU ukusheshisa

Kumenyezelwe mayelana nokuvula imithombo yenjini ye-SQL I-BlazingSQL, esebenzisa i-GPU ukusheshisa ukucutshungulwa kwedatha. I-BizingSQL akuyona i-DBMS egcwele, kodwa ibekwe njengenjini yokuhlaziya nokucubungula amasethi amakhulu edatha, aqhathaniseka nemisebenzi yayo I-Apache Spark. Ikhodi ibhalwe ngePython futhi vula ilayisensi ngaphansi kwe-Apache 2.0.

I-BizingSQL ifanele ukwenza imibuzo yokuhlaziya eyodwa kumasethi amakhulu wedatha (amashumi amagigabhayithi) agcinwe kumafomethi ethebula (isibonelo, amalogi, izibalo ze-NetFlow, njll.). I-BizingSQL ingaqalisa imibuzo kusuka kumafayela aluhlaza kufomethi ye-CSV ne-Apache Parquet asingathwe kunethiwekhi nezinhlelo zefayela lamafu njenge-HDSF ne-AWS S3, idlulisela ngokuqondile umphumela kumemori ye-GPU. Ngenxa yokufana kokusebenza ku-GPU kanye nokusetshenziswa kwememori yevidiyo esheshayo, imibuzo ku-BizingSQL iyenziwa kuze kube Izikhathi ze-20 ngokushesha kune-Apache Spark.

Ukuze usebenze nama-GPU, isethi eyakhiwe ngokubamba iqhaza kwe-NVIDIA iyasetshenziswa vula imitapo yolwazi RAPIDS, okukuvumela ukuthi udale ukucubungula idatha nezinhlelo zokusebenza ze-analytics ezisebenza ngokuphelele ohlangothini lwe-GPU (kuhlinzekwe ngu Python interface ukusebenzisa ama-primitives e-CUDA asezingeni eliphansi futhi ahambisane nezibalo).

I-BizingSQL inikeza amandla okusebenzisa i-SQL esikhundleni sama-API okucubungula idatha kuUDF (kwisisekelo Umcibisholo we-Apache) esetshenziswa ku-RAPIDS. I-BizingSQL isendlalelo esengeziwe esisebenza phezu kwe-cuDF futhi isebenzisa umtapo wezincwadi we-cuIO ukufunda idatha kudiski. Imibuzo ye-SQL ihunyushwa ibe izingcingo eziya emisebenzini ye-cuUDF, ekuvumela ukuthi ulayishe idatha ku-GPU futhi wenze imisebenzi yokuhlanganisa, ukuhlanganisa nokuhlunga kuyo. Ukudalwa kokucushwa okusabalalisiwe okuhlanganisa izinkulungwane zama-GPU kuyasekelwa.

I-BizingSQL ikwenza kube lula kakhulu ukusebenza ngedatha - esikhundleni samakhulu ezingcingo eziya emisebenzini ye-cuDF, ungasebenzisa umbuzo owodwa we-SQL. Ukusetshenziswa kwe-SQL kwenza kube lula ukuhlanganisa i-RAPIDS nezinhlelo ezikhona zokuhlaziya, ngaphandle kokubhala amaphrosesa athile futhi ngaphandle kokusebenzisa ukulayisha okuphakathi kwedatha ku-DBMS eyengeziwe, kodwa
kuyilapho kugcinwa ukuhambisana okugcwele nazo zonke izingxenye ze-RAPIDS, kuhunyushwa ukusebenza okukhona ku-SQL futhi kuhlinzeka ukusebenza ezingeni le-cuDF. Lokhu kubandakanya ukwesekwa kokuhlanganiswa namalabhulali XGBoost и kuML ukuxazulula izinkinga zezibalo nokufunda komshini.

Source: opennet.ru