2/3/09

Block diagram of a DBMS Storing Data: Disks and Files

Lecture 3 (R&G Chapter 9)
Query Optimization and Execution Relational Operators Files and Access Methods Buffer Management Disk Space Management Concurrency Control and Recovery

Yea, from the table of my memory Ill wipe away all trivial fond records. -- Shakespeare, Hamlet

DB

Disks, Memory, and Files

Disks and Files

Query Optimization and Execution Relational Operators Files and Access Methods Buffer Management Disk Space Management

DBMS stores information on disks.

Disks are a mechanical anachronism!

Major implications for DBMS design!

READ: transfer data from disk to main memory (RAM). WRITE: transfer data from RAM to disk. Both high-cost relative to memory references
Can/should plan carefully!

DB

Why Not Store Everything in Main Memory?

The Storage Hierarchy

Smaller, Faster Main memory (RAM) for currently used data. Disk for main database (secondary storage). Tapes for archive (tertiary storage). The role of Flash (SSD) still unclear

Costs too much. For ~$1000, PCConnection will sell you either
~80GB of RAM (unrealistic) ~400GB of Flash USB keys (unrealistic) ~180GB of Flash solid-state disk (serious) ~7.7TB of disk (serious)

Main memory is volatile.

Want data to persist between runs. (Obviously!)

Bigger, Slower
Source: Operating Systems Concepts 5th Edition

2/3/09

Jim Grays Storage Latency Analogy: How Far Away is the Data?
10 9 Andromeda Tape /Optical Robot Pluto 2,000 Years

Disks
Still the secondary storage device of choice. Main advantage over tape:
random access vs. sequential.

10 6 Disk

2 Years

Fixed unit of transfer

Read/write disk blocks or pages (8K)

Not random access (vs. RAM)

100 10 2 1 Memory On Board Cache On Chip Cache Registers Sacramento 1.5 hr This Building 10 min This Room My Head 1 min

Time to retrieve a block depends on location Relative placement of blocks on disk has major impact on DBMS performance!

Components of a Disk
Disk head Spindle Tracks

Accessing a Disk Page

Time to access (read/write) a disk block:
Sector

The platters spin (say, 120 rps). The arm assembly is moved in or out to position a head on a desired track. Tracks under heads make a cylinder (imaginary!). Only one head reads/ writes at any one time.
Block

seek time (moving arms to position disk head on track) rotational delay (waiting for block to rotate under head) transfer time (actually moving data to/from disk surface)

Arm movement

Platters

Seek time and rotational delay dominate.

Seek time varies from 0 to 10msec Rotational delay varies from 0 to 3msec Transfer rate around .02msec per 8K block

size is a multiple of sector size (which is fixed).

Arm assembly

Key to lower I/O cost: reduce seek/rotation delays! Hardware vs. software solutions?

Arranging Pages on Disk

`Next block concept:
blocks on same track, followed by blocks on same cylinder, followed by blocks on adjacent cylinder

Disk Space Management

Lowest layer of DBMS, manages space on disk Higher levels call upon this layer to:
allocate/de-allocate a page read/write a page

Blocks in a file should be arranged sequentially on disk (by `next), to minimize seek and rotational delay. For a sequential scan, pre-fetching several pages at a time is a big win!

Request for a sequence of pages best satisfied by pages stored sequentially on disk!
Responsibility of disk space manager. Higher levels dont know how this is done, or how free space is managed. Though they may make performance assumptions!
Hence disk space manager should do a decent job.

2/3/09

Context

Buffer Management in a DBMS

Page Requests from Higher Levels
Query Optimization and Execution Relational Operators Files and Access Methods Buffer Management Disk Space Management
copy of disk page free frame MAIN MEMORY DISK disk page DB
A

BUFFER POOL

DB

choice of frame dictated by replacement policy

Data must be in RAM for DBMS to operate on it! BufMgr hides the fact that not all data is in RAM

When a Page is Requested ...

Buffer pool information table contains: <frame#, pageid, pin_count, dirty> [Link] requested page is not in pool:
a. Choose a frame for replacement. Only un-pinned pages are candidates! b. If frame dirty, write current page to disk c. Read requested page into frame

More on Buffer Management

Requestor of page must eventually:
1. unpin it 2. indicate whether page was modified via dirty bit.

Page in pool may be requested many times,

a pin count is used. To pin a page: pin_count++ A page is a candidate for replacement iff pin count == 0 (unpinned)

[Link] the page and return its address.

If requests can be predicted (e.g., sequential scans) pages can be pre-fetched several pages at a time!

CC & recovery may do additional I/Os upon replacement.

Write-Ahead Log protocol; more later!

Buffer Replacement Policy

Frame is chosen for replacement by a replacement policy:
Least-recently-used (LRU), MRU, Clock,

LRU Replacement Policy

Least Recently Used (LRU)
(Frame pinned: in use, not available to replace) track time each frame last unpinned (end of use) replace the frame which has the earliest unpinned time

Policy can have big impact on #I/Os;

Depends on the access pattern.

Very common policy: intuitive and simple

Works well for repeated accesses to popular pages

Problem: Sequential flooding

LRU + repeated sequential scans. # buffer frames < # pages in file means each page request causes an I/O. Idea: MRU better in this scenario? Well see in HW1!

2/3/09

Clock Replacement Policy

D(1)

A(1) B(p)

DBMS vs. OS File System

OS does disk space & buffer mgmt: why not let OS manage these tasks? Buffer management in DBMS requires ability to:
pin a page in buffer pool, force a page to disk & order writes (important for implementing CC & recovery) adjust replacement policy, and pre-fetch pages based on access patterns in typical DB operations.

An approximation of LRU C(1) Arrange frames into a cycle, store one reference bit per frame
Can think of this as the 2nd chance bit

When pin count reduces to 0, turn on ref. bit When replacement necessary:
do for each page in cycle { if (pincount == 0 && ref bit is on) turn off ref bit; // 2nd chance else if (pincount == 0 && ref bit is off) choose this page for replacement; } until a page is chosen;

I/O typically done via lower-level OS interfaces

Avoid OS file cache Control write timing, prefetching

Context

Files of Records
Blocks are the interface for I/O, but Higher levels of DBMS operate on records, and files of records. FILE: A collection of pages, each containing a collection of records. Must support:
insert/delete/modify record fetch a particular record (specified using record id) scan all records (possibly with some conditions on the records to be retrieved)

Query Optimization and Execution Relational Operators Files and Access Methods Buffer Management Disk Space Management

DB

Typically implemented as multiple OS files

Or raw disk space

Unordered (Heap) Files

Collection of records in no particular order. As file shrinks/grows, disk pages (de)allocated To support record level operations, we must:
keep track of the pages in a file keep track of free space on pages keep track of the records on a page

Heap File Implemented as a List

Data Page Header Page Data Page

Data Page

Data Page

Full Pages

Data Page

Data Page

Pages with Free Space

There are many alternatives for keeping track of this.

Well consider 2

The header page id and Heap file name must be stored someplace.
Database catalog

Each page contains 2 `pointers plus data.

2/3/09

Heap File Using a Page Directory

Header Page Data Page 1 Data Page 2

Indexes (a sneak preview)

A Heap file allows us to retrieve records:

by specifying the rid, or by scanning all records sequentially

DIRECTORY

Data Page N

Sometimes, we want to retrieve records by specifying the values in one or more fields, e.g.,
Find all students in the CS department Find all students with a gpa > 3

The entry for a page can include the number of free bytes on the page. The directory is a collection of pages; linked list implementation is just one alternative.
Much smaller than linked list of all HF pages!

Indexes are file structures that enable us to answer such value-based queries efficiently.

Record Formats: Fixed Length

Record Formats: Variable Length

Two alternative formats (# fields is fixed):

F1 L1

F2 L2

F3 L3

F4 L4

F1

F2

F3

F4

Fields Delimited by Special Symbols

F1 F2 F3 F4

Base address (B)

Address = B+L1+L2

Information about field types same for all records in a file; stored in system catalogs. Finding ith field done via arithmetic.

Array of Field Offsets Second offers direct access to ith field, efficient storage of nulls (special dont know value); small directory overhead.

Page Formats: Fixed Length Records

Slot 1 Slot 2 Slot 1 Slot 2

Page Formats: Variable Length Records

Rid = (i,N) Rid = (i,2) Rid = (i,1) Page i

...
Slot N

Free Space Slot N Slot M N

...

1 . . . 0 1 1M number of records M ... 3 2 1 UNPACKED, BITMAP number of slots 20 N 16 ... 2 24

N
1# slots

PACKED

Record id = <page id, slot #>. In first alternative, moving records for free space management changes rid; may not be acceptable.

Can move records on page without changing rid; so, attractive for fixed-length records too.

SLOT DIRECTORY

Pointer to start of free space

2/3/09

System Catalogs
For each relation:
name, file location, file structure (e.g., Heap file) attribute name and type, for each attribute index name, for each index integrity constraints

Attr_Cat(attr_name, rel_name, type, position)

attr_name attr_name rel_name type position sid name login age gpa fid fname sal rel_name Attribute_Cat Attribute_Cat Attribute_Cat Attribute_Cat Students Students Students Students Students Faculty Faculty Faculty type string string string integer string string string integer real string string real position 1 2 3 4 1 2 3 4 5 1 2 3

For each index:

structure (e.g., B+ tree) and search key fields

For each view:

view name and definition

Plus statistics, authorization, buffer pool size, etc.

Catalogs

are themselves stored as relations!

pg_attribute

Summary
Disks provide cheap, non-volatile storage.
Better random access than tape, worse than RAM Arrange data to minimize seek and rotation delays.
Depends on workload!

Buffer manager brings pages into RAM.

Page pinned in RAM until released by requestor. Dirty pages written to disk when frame replaced (sometime after requestor unpins the page). Choice of frame to replace based on replacement policy. Tries to pre-fetch several pages at a time.

Summary (Contd.)
DBMS vs. OS File Support
DBMS needs non-default features Careful timing of writes, control over prefetch

Summary (Contd.)
DBMS File tracks collection of pages, records within each.
Pages with free space identified using linked list or directory structure

Variable length record format

Direct access to ith field and null values.

Slotted page format

Variable length records and intra-page reorg

Indexes support efficient retrieval of records based on the values in some fields. Catalog relations store information about relations, indexes and views.