Nambala ya injini ya BlazingSQL SQL yotseguka, pogwiritsa ntchito GPU kuti ipititse patsogolo

Adalengezedwa za kutsegula magwero a injini ya SQL BlazingSQL, yomwe imagwiritsa ntchito GPU kuti ifulumizitse kukonza deta. BlazingSQL si DBMS yathunthu, koma ili ngati injini yowunikira ndikukonza ma data akulu, ofanana ndi ntchito zake. Apache Spark. Khodiyo idalembedwa mu Python ndi ndi lotseguka zololedwa pansi pa Apache 2.0.

BlazingSQL ndiyoyenera kuchita mafunso amodzi owunikira pamaseti akulu akulu (makumi a ma gigabytes) osungidwa mumitundu yama tabular (mwachitsanzo, zipika, NetFlow statistics, etc.). BlazingSQL imatha kuyendetsa mafunso kuchokera pamafayilo aiwisi mu CSV ndi ma Apache Parquet omwe amakhala pamanetiweki ndi makina amafayilo amtambo monga HDSF ndi AWS S3, kusamutsa mwachindunji zotsatira ku kukumbukira kwa GPU. Chifukwa cha kufanana kwa magwiridwe antchito mu GPU komanso kugwiritsa ntchito kukumbukira kwamakanema mwachangu, mafunso a BlazingSQL amaperekedwa mochepera kuposa Nthawi 20 mwachangu kuposa Apache Spark.

Nambala ya injini ya BlazingSQL SQL yotseguka, pogwiritsa ntchito GPU kuti ipititse patsogolo

Kuti mugwire ntchito ndi ma GPU, seti yopangidwa ndi NVIDIA imagwiritsidwa ntchito tsegulani malaibulale KUDWALITSA, zomwe zimakupatsani mwayi wopanga ma data ndi ma analytics omwe amayendera mbali zonse za GPU (zoperekedwa ndi Python mawonekedwe kugwiritsa ntchito zoyambira zazing'ono za CUDA ndikufananitsa mawerengedwe).

BlazingSQL imapereka mwayi wogwiritsa ntchito SQL m'malo mwa ma data processing API kuUDF (pa base Apache Arrow) amagwiritsidwa ntchito mu RAPIDS. BlazingSQL ndi gawo lowonjezera lomwe limayenda pamwamba pa cuDF ndipo limagwiritsa ntchito laibulale ya cuIO kuti iwerenge deta kuchokera pa disk. Mafunso a SQL amamasuliridwa kukhala mafoni ku ntchito za cuUDF, zomwe zimakupatsani mwayi wotsitsa deta mu GPU ndikuchita kuphatikiza, kuphatikizira ndi kusefa pamenepo. Kupanga masinthidwe ogawidwa omwe amatenga masauzande a ma GPU kumathandizidwa.

BlazingSQL imathandizira kwambiri kugwira ntchito ndi deta - m'malo mwa mazana a mafoni ku ntchito za cuDF, mutha kugwiritsa ntchito funso limodzi la SQL. Kugwiritsa ntchito SQL kumathandizira kuphatikizira RAPIDS ndi machitidwe omwe alipo, osalemba mapurosesa apadera komanso osagwiritsa ntchito kutsitsa kwapakatikati kwa data mu DBMS yowonjezera, koma
ndikusunga kuyanjana kwathunthu ndi magawo onse a RAPIDS, kumasulira magwiridwe antchito mu SQL ndikupereka magwiridwe antchito pamlingo wa cuDF. Izi zikuphatikiza kuthandizira kuphatikiza ndi malaibulale XGBoost ΠΈ kuml kuthetsa mavuto a analytics ndi kuphunzira makina.

Source: opennet.ru

Kuwonjezera ndemanga