Storage and Indexes
Introduction to DatabasesComputer Science 557
Instructor: Joe BockhorstUniversity of Wisconsin - Milwaukee
Announcements
• Any problems logging in to course accounts?
• Use grid3, grid5 or weise for course work– Send bugs on weise to [email protected]
• Reading Assignment: Chapter 14 in the textbook• Program 1 assigned today, due next Friday
Hard Disks• DBMS typically store information on “hard” disks
• I/O operations (“Read” and “Write”) are costly – Should be planned carefully
• A block is amount of data transferred in one operation (example block size ~ 1024 bytes)
DiskMain Memory
(RAM)
Read
Write
Hard Drive Glossary
• Disks are divided into concentric circular trackson each disk surface.– Track capacities vary typically from ~ 4 to 50 Kbytes
• The division of a track into sectors is hard-coded on the disk surface and cannot be changed
• A block is an integer number of sectors– The block size B is fixed for each system.
• Typical block sizes range from B=512 bytes to B=4096 bytes.– Whole blocks are transferred between disk and main memory
Accessing a Disk Block
• to Read or Write block give block addr to disk controller – Hardware block address – cylinder #, surface #, block #– Logical block addressing allows higher levels to refer to
hardware block address using a block_id• Seek time – move head to correct cylinder (~10ms)• Rotation time – rotate start of block under head
– @ 15000 rpm, average rotation time is 4ms• Transfer time – transfer entire block• Accessing consecutive blocks only need to pay the seek
and rotation time once• Compare to typical main memory access times which
are measured in micro (10-6) or nano (10-9) seconds
Coming soon: Solid State “Disks”?
+ Random access devices eliminate seek times
+ faster startup- $$$$ much more
expensive than hard disks - ($8 / GB vs $0.25 / GB)
- Capacities are smaller
We will assume hard disk storage inthis course
Why not store DB in main memory?
• $$$$– cost of RAM is > 100 X hard drive cost
• main memory is volatile• 32 bit addressing
Managing the Hard Disk
Query Optimization
Relational Operators
Files and Access Methods
Buffer Management
Disk Space Management
The DSM provides an abstraction of the block as a unit of data
DSM interface includes commands to read and write block commands
I/O requests
Operations Supported by DSM
• allocate_blocks(num_blocks)– Add blocks to DB
• deallocate block(blockID)– Remove block from DB
• write_block(blockID, blockPtr)– Write block to disk
• read_block(blockID, blockPtr)– Read block from disk
Managing the Hard Disk
Disk Space Manager
1yes
2 3block IDyes yesallocated?
N-1 Nno no
4 5no no
792 793 794 * ** *hardware addr
6no
7no
* *
Example: Managing the Hard Disk
allocate_blocks(3)write_block(5,data)read_block(5)deallocate block(2)
Disk Space Manager
1yes
2 3block IDyes yesallocated?
N-1 Nno no
4 5no no
792 793 794 * ** *hardware addr
6no
7no
* *
allocate_blocks(3)write_block(5,data)read_block(5)deallocate block(2)
Example: Managing the Hard Disk
Disk Space Manager
1yes
2 3block IDyes yesallocated?
N-1 Nno no
4 5yes yes
792 793 794 * *902 903hardware addr
6yes
7no
904 *
//allocate three consecutive blocks
allocate_blocks(3)write_block(5,data)read_block(5)deallocate block(2)
Example: Managing the Hard Disk
Disk Space Manager
1yes
2 3block IDyes yesallocated?
N-1 Nno no
4 5yes yes
792 793 794 * *902 903hardware addr
6yes
7no
904 *
//write blockID 5 to disk
write_block(903,data)
allocate_blocks(3)write_block(5,data)read_block(5)deallocate block(2)
Example: Managing the Hard Disk
Disk Space Manager
1yes
2 3block IDyes yesallocated?
N-1 Nno no
4 5yes yes
792 793 794 * *902 903hardware addr
6yes
7no
904 *
// read blockID 5 to buffer
read_block(903)
allocate_blocks(3)write_block(5,data)read_block(5)deallocate block(2)
Example: Managing the Hard Disk
Disk Space Manager
1yes
2 3block IDno yesallocated?
N-1 Nno no
4 5yes yes
792 * 794 * *902 903hardware addr
6yes
7no
904 *
// read blockID 5 to buffer
Buffer Management
• Responsible for managing region of main memory called the buffer pool
• MM pages are called frames (slots that can hold one block)
• Higher levels of the DBMS need not worry if the page is in memory or not... Just ask for it.
Query Optimization
Relational Operators
Files and Access Methods
Disk Space Management
Buffer Management
Buffer Manager Operations
• add_blocks_to_DB(num_blocks)– add new blocks to DB
• delete_block_from_DB(block_id)– delete block from the DB
• pin_block(block_id)– bring block from disk to buffer pool if not in BP– increment pin count for block
• unpin_block(block_id)– decrement pin count for block
• mark_dirty(block_id)
• Buffer Manager maintains for each frame– pin count– dirty bit
Buffer Manager Example
buffer poolwith M frames1 2 3 M-1 M
- - - - -
0 0 0 0 0
no no no no no
block IDpin count
dirty
initial state ofbuffer manager
1 2 3 N-1 N
Buffer Manager Example
buffer poolwith M frames1 2 3 M-1 M
76 22 - - -
2 1 0 0 0
no no no no no
block IDpin count
dirty
initial state ofbuffer manager
draw on whiteboard
1 2 3 N-1 N