BlazingSQL SQL injini kodhi yakavhurika, uchishandisa GPU yekumhanyisa

Kuziviswa nezve kuvhura masosi einjini yeSQL BlazingSQL, iyo inoshandisa GPU kukurumidza kugadzirisa data. BlazingSQL haisi DBMS yakazara-yakazara, asi yakamisikidzwa seinjini yekuongorora uye kugadzirisa makuru data seti, anofananidzwa mumabasa ayo Apache spark. Iyo kodhi yakanyorwa muPython uye kuvhura ine rezinesi pasi peApache 2.0.

BlazingSQL inokodzera kuita imwechete yekuongorora mibvunzo pane yakakura data seti (makumi emagigabytes) akachengetwa mumatabula mafomati (semuenzaniso, matanda, NetFlow nhamba, nezvimwewo). BlazingSQL inogona kumhanyisa mibvunzo kubva kumafuta akaomeswa muCSV neApache Parquet mafomati akabatwa panetiweki uye gore faira masisitimu seHDSF neAWS S3, ichiendesa zvakananga mhedzisiro kuGPU ndangariro. Nekuda kwekufananidza kwekushanda muGPU uye kushandiswa kwekukurumidza vhidhiyo ndangariro, BlazingSQL mibvunzo inoita isingasviki. 20 nguva nekukurumidza kupfuura Apache Spark.

Kushanda nemaGPU, seti yakagadzirwa nekutora chikamu kweNVIDIA inoshandiswa kuvhura maraibhurari RAPIDS, iyo inokutendera iwe kuti ugadzire data kugadzirisa uye analytics maapplication anomhanya zvachose padivi reGPU (yakapihwa na Python interface kushandisa yakaderera-level CUDA primitives uye parallelize kuverenga).

BlazingSQL inopa kugona kushandisa SQL pachinzvimbo chekugadzirisa data APIs cuUDF (pane base Apache Arrow) inoshandiswa muRAPIDS. BlazingSQL imwe yekuwedzera iyo inomhanya pamusoro pecuDF uye inoshandisa cuIO raibhurari kuverenga data kubva kudhisiki. Mibvunzo yeSQL inoshandurirwa kuita mafoni kumabasa ecuUDF, ayo anobvumidza iwe kurodha data muGPU uye kuita kubatanidza, kuunganidza uye kusefa mashandiro pairi. Kugadzirwa kwezvirongwa zvakagoverwa zvinotora zviuru zveGPU zvinotsigirwa.

BlazingSQL inorerutsa zvakanyanya kushanda nedata - pachinzvimbo chemazana ekufona kumabasa ecuDF, unogona kushandisa imwe SQL mubvunzo. Iko kushandiswa kweSQL kunoita kuti zvikwanise kubatanidza RAPIDS nearipo analytics masisitimu, pasina kunyora chaiwo ma processor uye pasina kushandisa yepakati kurodha data mune imwe DBMS, asi.
uku uchichengetedza kuenderana kwakazara nezvikamu zvese zveRAPIDS, kushandura mashandiro aripo muSQL uye kupa mashandiro padanho recuDF. Izvi zvinosanganisira rutsigiro rwekubatanidza nemaraibhurari XGBoost и cuML yekugadzirisa matambudziko eanalytics uye kudzidza muchina.

Source: opennet.ru