SQL Server 2019 querying 1 trillion rows in 100 seconds

preview_player
Показать описание
We recently performed a performance test on a SQL Server 2019 instance running on a Windows Server 2016 running on an 8-Socket (224 cores, 12TB RAM and 200TB+ SSD storage) Lenovo Server (ThinkSystem SR950) using Intel Cascade Lake processors. We generated data and loaded a single table (LINEITEM, TPCH database schema with 145TB+ raw data). This table had 54TB compressed data with 1 Trillion+ rows. Q1 query (as defined by TPCH) which scans the whole table and selects nearly all the rows for computation was run in both cold (all data read from storage) and warm (data in memory) scenario. SQL Server 2019 was able to provide unparalleled performance processing over a trillion rows with the query completing in under 2 mins (107 secs) in warm cache and under 4 mins (238 secs) with all data read from storage. For the warm cache, this translates to processing over 6 billion rows/sec and read throughput of around 50GB/sec. You can find the Azure Data Studio Notebooks that showcase both scenarios on GitHub.
Рекомендации по теме
Комментарии
Автор

8 CPU Sockets? Thats Insane! I love it!

johnsa
Автор

how to maximize processor performance so that it can be used 100%
on my query running slow but usage can't reach 100%

blackak
Автор

Me reading memory: 12gb only??? Wtf how. Jump back and realized the t (tb). I Got the best laugh flash in a long time.

a_red_panda
Автор

Hi i have one question . How to so select query for 100 million recordset from mssql with rapidly. It's possible?

Common_man_views
Автор

Cool, right about 0.00001% of the world's businesses will afford the licensing costs so if this is supposed to be a flex I am not impressed.

DanielWillen