BlazingSQL SQL lambar injin buɗewa, ta amfani da GPU don haɓakawa

An sanar game da buɗe tushen injin SQL BlazingSQL, wanda ke amfani da GPU don hanzarta sarrafa bayanai. BlazingSQL ba cikakken DBMS ba ne, amma an sanya shi azaman injiniya don nazari da sarrafa manyan saitin bayanai, kwatankwacin a cikin ayyukansa zuwa Apache Spark. An rubuta lambar a Python kuma a bude lasisi a ƙarƙashin Apache 2.0.

BlazingSQL ya dace don yin tambayoyin bincike guda ɗaya akan manyan saitin bayanai (dubun gigabytes) da aka adana a cikin tsarin tambura (misali, rajistan ayyukan, ƙididdiga na NetFlow, da sauransu). BlazingSQL na iya gudanar da tambayoyi daga fayilolin da aka yi amfani da su a cikin CSV da Apache Parquet Formats wanda aka shirya akan hanyar sadarwa da tsarin fayil na girgije kamar HDSF da AWS S3, suna canja wurin sakamakon kai tsaye zuwa ƙwaƙwalwar GPU. Godiya ga daidaituwar ayyuka a cikin GPU da amfani da ƙwaƙwalwar bidiyo mai sauri, tambayoyin BlazingSQL suna aiwatar da ƙasa da ƙasa. 20 sau sauri fiye da Apache Spark.

Don aiki tare da GPUs, ana amfani da saitin da aka haɓaka tare da sa hannun NVIDIA bude dakunan karatu SANARWA, wanda ke ba ka damar ƙirƙirar bayanan sarrafa bayanai da aikace-aikacen nazari waɗanda ke gudana gaba ɗaya a gefen GPU (wanda aka bayar ta Python interface don amfani da ƙananan matakan CUDA na farko da daidaita lissafi).

BlazingSQL yana ba da damar yin amfani da SQL maimakon API ɗin sarrafa bayanai kuUDF (na tushe Kibiya Apache) ana amfani dashi a cikin RAPIDS. BlazingSQL ƙarin Layer ne wanda ke gudana a saman cuDF kuma yana amfani da ɗakin karatu na cuIO don karanta bayanai daga faifai. Ana fassara tambayoyin SQL zuwa kira zuwa ayyukan cuUDF, waɗanda ke ba ku damar loda bayanai a cikin GPU da yin ayyukan haɗaka, tarawa da tacewa akansa. Ƙirƙirar jeri mai rarrabawa wanda ya kai dubunnan GPUs ana tallafawa.

BlazingSQL yana sauƙaƙa aiki da bayanai sosai - maimakon ɗaruruwan kira zuwa ayyukan cuDF, zaku iya amfani da tambayar SQL ɗaya. Yin amfani da SQL yana ba da damar haɗa RAPIDS tare da tsarin nazari na yanzu, ba tare da rubuta takamaiman na'urori ba kuma ba tare da yin amfani da matsakaicin nauyin bayanai a cikin ƙarin DBMS ba, amma
yayin da ake kiyaye cikakkiyar dacewa tare da duk sassan RAPIDS, fassara ayyukan da ake dasu zuwa SQL da samar da aiki a matakin cuDF. Wannan ya haɗa da goyan baya don haɗin kai tare da ɗakunan karatu. XGBoost и kuML don magance matsalolin nazari da koyon inji.

source: budenet.ru