2018-01-10

What are key design goals when building a DBMS?

support recovery from failures.

• storage failure
• End user failure/aborts.
• server failure/software, power, catastrophe (distributed locations)...
  Support concurrency
• give users/programmers illusion of being alone. (transactions)
  supporting high performance queries/updates indexing is one solution.
  Support "Big Data"
• transactional system (data expires and then archived to anal sys),
• analytic systems (data don't get expires)
• how big is "big"? Depends GBs -> PBs
  Support data independence
• the form of data is independent from computer systems.
• typically done via a declarative programming language (SQL)+optimization algorithms that implement that.

What does a model DBMS arch look like?

From top to bottom
programming interface: more like a compiler, e.g., SQL->C ; -----> Logging & Recovery Mgr, also ----> concurrency control mgr
|
logical Optimizer: better ways of implement it (SQL) -----> catlog:
|
----------->catlog: stores metadata (data about the data) --- you might use a minidatabase to manage the catlog, SQLLite
|
physical optimizer: attach algorithms to the implementation, e.g., this type of join...choose the algorithm. ---> catlog
|
query execution engine --->catlog, also ---> Logging & Recovery Mgr
|
------------> Logging & Recovery Mgr <------>concurrency control mgr
|
index/file/record mgr typed bits ---> catlog, also ----> Logging & Recovery Mgr, also -----> concurrency control mgr
|

Buffer manager: untype bits 1. for high performance 2. Strong interaction between concurrency control and system recovery ----> concurrency control mgr
|
Persist. Storage: DBMS will often manage the persist storage at a very low level (the OS may not be reliable).

logging & recovery, concurrency control mgr <-> most nasty part, but ignored in our project. (will covered in the quiz.)