Re:Scaling Relational SQL Databases (2020)
by mamcx on Wednesday June 16, 2021 @10:38AM
(#27529016)
In other words: You can't add many workers to a linear job and expect the task is done in SizeOfJob/Workers.
Or: If you put many car lanes that go into one, you will get contention.
This will manifest in many ways.
But you "can" if along ALL the pipeline you have a way to coordinate the task!. However, if componentA is paradigmA and componentB is paradigmB you will get trouble.
---
In practical term: Wanna kill your database? Open N connections launched by thousand of actors/async/threads. You need to instead use a pool, coordinate which jobs are fast to done or slow to complete.
AND
You need to architect the code to exploit both what this paradigm is, + what that make happy the DB (typical: NOT do sql + n, think in batches, group operations, etc)
This is the part you don't get much help not matter what you use.
Or: If you put many car lanes that go into one, you will get contention.
This will manifest in many ways.
But you "can" if along ALL the pipeline you have a way to coordinate the task!. However, if componentA is paradigmA and componentB is paradigmB you will get trouble.
---
In practical term: Wanna kill your database? Open N connections launched by thousand of actors/async/threads. You need to instead use a pool, coordinate which jobs are fast to done or slow to complete.
AND
You need to architect the code to exploit both what this paradigm is, + what that make happy the DB (typical: NOT do sql + n, think in batches, group operations, etc)
This is the part you don't get much help not matter what you use.