the idea for today is to cause #sqlserver [total server memory] to exceed [target server memory] by several gb with a rowstore-only workload and *then* read columnstore segments into buffer pool (and column store object pool).
while there are no columnstore related pages in bpool and columnstore object pool does not exist, the impossibly-high [target pages] value means almost no SQLOS free memory on the system.
initializing column store object pool brings [target pages] to a value well under [target server memory (kb)] - though, note [target pages] is measured in 8kb pages.
the result is a rapid increase of SQLOS free memory as pages from bpool are freed. but if the rate of freeing bpool pages is less than the rate of reading new pages into bpool while free memory is growing… #sqlserver [total server memory] will try to grow even more past target.
that’s the theory.
let’s see if I can realize it.
let’s see if I can realize it.
https://twitter.com/sqL_handLe/status/1481290888374632450
• • •
Missing some Tweet in this thread? You can try to
force a refresh
