With Aurora, the same operations of placing a page into the buffer pool, writing to the redo log, and modifying the page in the buffer pool remain (View Highlight)
with each redo log record representing the changes to be made to a page. The major difference is that Aurora doesn’t issue checkpoints, ever. Instead, the only data that is written to the Aurora storage volume is the redo log, which is very small in size as compared to a full page. This means that all the disk I/O that is normally generated by writing dirty 8 KB or 16 KB pages to disk is eliminated. Instead, pages are materialized periodically on the storage nodes and even on demand in some cases (View Highlight)
With Aurora, reads are handled in same way as native MySQL or PostgreSQL—data is read from the storage volume by page as needed to populate the buffer pool, and SELECT queries are run against those pages inside the buffer pool (View Highlight)
