Memory Management in MySQL and Drizzle

How does the database server use memory? Where? What are the scaling implications of how we allocate, use and free memory in the database server? Can we scale to many CPU cores effectively? Do alternate malloc libraries really help? Why? What does MySQL do? What does Drizzle do (and what have we changed?). These questions (and more) will be answered in this session.

We’ll cover a variety of caches in the database server and how they’re allocated and used (query cache, key cache, table cache etc). We’ll delve into the server internals and go over “What is a MEM_ROOT?”, how they’re implemented, the impact of them over plain malloc() and other types of allocators. What impact does this have for tools such as valgrind? Are we getting speed at the expense of the usuability of tools such as valgrind? Are we using more memory than we need to? Is this a space vs speed tradeoff?

What do various storage engines do? Are they any better/worse than the SQL server?

I’ll present research and experience on the MySQL and Drizzle database servers. I’ll present new work on Drizzle (a whole bunch of which is in progress and yet to be completed) and plans of how to get to a more efficient database server as well as one that’s easier to write correct code for.