Khoutu ea enjine ea BlazingSQL SQL e butsoe, e sebelisa GPU bakeng sa ho potlakisa

Tsebisitsoe mabapi le ho bula mehloli ea enjene ea SQL Blazingsql., e sebelisang GPU ho potlakisa ts'ebetso ea data. BlazingSQL ha se DBMS e felletseng, empa e behiloe joalo ka enjene bakeng sa ho sekaseka le ho sebetsana le li-sete tse kholo tsa data, tse bapisoang le mesebetsing ea eona. Apache Spark. Khoutu e ngotsoe ka Python le bula e nang le tumello tlas'a Apache 2.0.

BlazingSQL e loketse ho etsa lipotso tsa tlhahlobo e le 'ngoe ho li-data tse kholo (li-gigabyte tse mashome) tse bolokiloeng ka liforomo tsa tabular (mohlala, li-logs, lipalo-palo tsa NetFlow, joalo-joalo). BlazingSQL e ka tsamaisa lipotso ho tsoa lifaeleng tse tala ka liforomo tsa CSV le Apache Parquet tse tsamaisoang ke marang-rang le lits'ebetso tsa lifaele tsa maru joalo ka HDSF le AWS S3, e fetisetsa sephetho ka kotloloho mohopolong oa GPU. Ka lebaka la ho ts'oana ha ts'ebetso ho GPU le ts'ebeliso ea memori ea video e potlakileng, lipotso tsa BlazingSQL li etsoa ka tlase ho Linako tsa 20 kapele ho feta Apache Spark.

Ho sebetsa le li-GPU, ho sebelisoa sete e ntlafalitsoeng ka ho nka karolo ha NVIDIA bula lilaeborari RAPIDS, e u lumellang hore u thehe lisebelisoa tsa ts'ebetso ea data le li-analytics tse tsamaeang ka ho feletseng ka lehlakoreng la GPU (e fanoeng ke Python interface ho sebelisa li-primitives tsa boemo bo tlase ba CUDA le ho bapisa lipalo).

BlazingSQL e fana ka bokhoni ba ho sebelisa SQL ho fapana le li-API tsa ts'ebetso ea data hoUDF (ka motheo Motsu oa Apache) e sebelisoang ho RAPIDS. BlazingSQL ke lera le eketsehileng le tsamaeang ka holim'a cuDF 'me le sebelisa laebrari ea cuIO ho bala data ho tswa ho disk. Lipotso tsa SQL li fetoleloa e le mehala ho mesebetsi ea cuUDF, e u lumellang ho kenya data ho GPU le ho etsa ts'ebetso ea ho kopanya, ho kopanya le ho sefa ho eona. Theho ea litlhophiso tse ajoang tse nkang likete tsa li-GPU lia tšehetsoa.

BlazingSQL e nolofatsa haholo ho sebetsa ka data - sebakeng sa mehala e makholo ho mesebetsi ea cuDF, o ka sebelisa potso e le 'ngoe ea SQL. Tšebeliso ea SQL e etsa hore ho khonehe ho kopanya RAPIDS le litsamaiso tse teng tsa analytics, ntle le ho ngola li-processor tse khethehileng le ntle le ho sebelisa mokhoa oa ho kenya data ho DBMS e eketsehileng, empa
ha re ntse re boloka tumellano e felletseng le likarolo tsohle tsa RAPIDS, ho fetolela ts'ebetso e teng ho SQL le ho fana ka ts'ebetso maemong a cuDF. Sena se kenyelletsa tšehetso ea ho hokahana le lilaebrari XGBoost и cuML bakeng sa ho rarolla mathata a analytics le ho ithuta ka mochini.

Source: opennet.ru