CMU Advanced Database Systems - (Spring 2019)

[Sunt-ing]

https://www.youtube.com/playlist?list=PLSE8ODhjZXja7K1hjZ01UTVDnGQdx5v5U

Since I have basic knowledge about database, I may skip the basic course about database.

[Sunt-ing]

Lecture 1: In-Memory Databases

• The history of DBMSs development is about dealing with the limitations of hardware
• In a disk-based system, only approximately 7% of instructions are done on actual work. The majority of
  the DBMS’s instructions time are in managing three of its key components: (1) buffer pool, (2) concurrency
  control, (3) logging/recovery
• Concurrency Control: In-memory DBMS can store locking information about each tuple together with its data. This is because the cost of a transaction acquiring a lock is the same as accessing data. Contrast this with disk-oriented DBMSs where locks are physically stored separate from their tuples because the tuples may get swapped out to disk.
• Indexes: In-memory DBMSs will not log index updates. Instead, the system will rebuild the indexes upon restart
  when it loads the database back into memory. This avoids the runtime overhead of logging updates to
  indexes during transaction execution.
• Logging and Recovery: more lightweight logging schemes (e.g., only store redo information) is used. For example, since there are no “dirty pages”, the DBMS does not need to maintain LSNs. In-memory DBMSs still takes checkpoints to reduce the amount of log that the system has to replay during recovery.