storage
- DBMS always target maximizing sequential access (on non-volatile memory)
- design goal of DBMS is managing DB that exceeds the amount of memory available
- not use
mmap in DBMS due to correctness/performance
pages
- fixed-size block of data (1-16KB)
- page layout
- slotted pages: maps slots to offsets
storage layout
- tuple-oriented storage
- store tuples on disk using the slotted-page scheme
- check page directory to find page position in disk, fetch page to memory, find free slot in this page and update
- log-structured storage
- maintains a log that records changes to tuples
- DBMS applies changes to an in-memory data structure (MemTable) and writes out sequentially to disk (SSTable)
- SSTables are immutable -> append-only storage
- use merging algorithm to combine SSTables into larger one (taking most recent changes for each tuple)
storage models
- N-Array storage model (NSM)
- decomposition storage model (DSM)
memory
- move data back and forth between disk and memory with Buffer Pool Manager
buffer pool
- is in-memory cache (write-back) of pages
- is array of fixed-size frames
- page directory maps page IDs to page locations in DB files. all changes must be recorded on disk
- page table maps page IDs to a copy of pages in buffer pool frames
- DBMS can maintain multiple buffer pools with (local) tailored policies
locks/latches
- lock is a higher-level, logical primitive that protects the contents of a DB from other transactions
- latch is a low-level protection primitive that DBMS uses for critical sections in internal data structures
buffer replacement policies
- LRU
- CLOCK
- similar to LRU without a separate timestamp per page (use reference bit - 1 is page is accessed)
- LRU/CLOCK are susceptible to sequential flooding
- LRU-K
- ARC: adaptive replacement cache
disk IO, OS cache
- DBMS maintains internal queues to track page read/write