CS186-L3: Disk, Buffers, Files I

big picture

sql client -> DBMS -> database 🤓

DBMS

• parsing & optimization
  • 执行SQL语句时，DBMS需要解析SQL语句，并将其转换为执行计划。优化器会根据统计信息、查询模式、索引等因素，选择最优的执行计划。
• relational operators
  • 处理数据流 or 关系运算符？
• files and index management
• buffer management
• disk space management

事实上纵向还有两个模块：concurrency control和recovery。

省流：从RAM & DISK获取数据非常慢， 相对于CPU

Disk

注意sector, disk head， 其中后者似乎只能单次读写

access time

flash SSD

注意：

read很快，随着数据变大，可以预测

write很慢，slower for random，写入放大

disk space management

block level storage

block: unit of transfer for disk read/write (64~128KB in 2018)

page: a common synonym for block, in some contexts, it means in RAM

implementation

• talk to hardware directly 🤔
• use file system (FS) 😀
  • always remember: next is fast

files and representation

important! 😎

• tables stored in files
  • consist of pages
    • pages contain a collection of records

DB files

• unordered heap files
  • DLL是一种简单的实现heap file的方案，但是insert效率不高 😞
  • better: a page directory 注意header page is SLL

layout of a page

• page header
  • nums of records
  • free space
  • maybe a next/last pointer
  • bitmaps, slot table (what is that? 🤔)

注意，是否records定长以及是否有free space决定了page layout

fixed-length packed records

now take a look at a fixed length records, packed page:

fixed-length unpacked records

same as before, but with unpacked records:

variable-length records

records can have different lengths 🤯

• header -> footer
• footer has a slot directory (read from right to left, has a pointer to the start of the free space)
• slot directory save slot, each slot has a pointer to the start of the record and the length of the record
  • growing slot directory, 前后夹击
  • 

layout of records

• each record has a fixed type
• system catalog stores the SCHEMA
  • no need to store the type of records
  • catalog just a table

注意，以下讨论在字段（field）级别，不是record级别。

fixed-length records

类比数组，注意null存在就是空着，不是很compact

variable-length records

• 冗余配置按照定长处理 （padding）
• 类比CSV，每行记录不同长度，用分隔符分隔
• a record header way