Structure of a DBMS• A typical DBMS has a

layered architecture.

• The figure does not show the concurrency control and recovery components.

• This is one of several possible architectures; each system has its own variations.

Query Optimizationand Execution

Relational Operators

Files and Access Methods

Buffer Management

Disk Space Management

DB

These layersmust considerconcurrencycontrol andrecovery

Query Parser/Decomp.

Data Storage

DBMS deals with a very large amount of data.

• How does a computer system store and manage very large volumes of data?

• What representations and data structures best support efficient manipulations of this data?

Storage HierarchyCPU

CACHE

MAIN MEMORY

HARD DISK (RAID)

OPTICAL DISK

TAPE

KB

MB

GB

TB

TB

Redundant Array Inexpensive Disk

RAID: A number of disks is organized and appears to be a single one to the OS

• Aggregate disk capacity

• Increase I/O throughput

• Fault-tolerant (hot swapping)

PC

RAID System

Each database contains a number of tables Each table contains a number of records Each record has a unique identifier called a

record id, or rid. Records can be stored in files based on the underlying

OS. Records can be stored directly to disk blocks bypassing

file systems (raw devices).

Storing databases/tables/records

Disk Manager

Buffer Manager

Why raw devices?• Differences in OS support: portability

issues• Some limitations, e.g., files can’t span

disks.• Performance

Higher Layer

Lowest level in DBMS software architectureOS files disksor

Access Manager

Arrange records in a page.

An overview at 30000-feet

High

Disk Manager

Buffer Manager

Higher Layer

OS files disksor

Access Manager

Record 1Record 2

:::::Record n1

CreateTable(DB, tName, field[])ReadRecord(DB, tName, rID)WriteRecord(DB, tName, rID, r)DeleteRecord(DB, tName, rID)Append(DB, tName, record)

record-level operations

Page 1Page 2

:::::Page mAllocatePage()

DeletePage (pID)WritePage (pID, page)ReadRecord(pID)

page-level operations

Mapping between database-table-record

and page

Db1:T1Record 1Record 2

Record n2

Dbi:Tj

:: :::::

Mapping between page and devices

HD Optical Tape

Disk Manager• Higher levels call upon this layer to:

– allocate/de-allocate a page on disk– read/write a page (or a block or a unit of disk

retrieval)

Disk Manager

Buffer Manager

Higher Layer

OS files disksor

Access Manager

Lowest level in DBMS software architecture

Arrange records in a page.

Page 1Page 2Page 3Page 4

:::::Page n

These pages are physically located in different devices (RAID, optical, and/or tape)

Buffer Manager

DB

MAIN MEMORY

DISK

disk page

free frame

Page Requests from Higher LevelsBUFFER POOL

choice of frame dictatedby replacement policy

Size of a frame equal to size of a disk pageTwo variables associated with each frame/page:

• Pin_count: Number of current users of the page• Dirty: whether the page has been modified since it has been

brought into the buffer pool.

Buffer ManagementWhen a Page is Requested ...

• If a requested page is not in the buffer pool:– Choose a frame for replacement– If the frame is dirty, write it to disk– Read the requested page into the chosen frame

• Pin the page and return its address to the requester.

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

Pinning: Incrementing a pin_count.Unpinning: Release the page and the pin_count is decremented.

Buffer Replacement Policy• Frame is chosen for replacement by a

replacement policy:– Least-recently-used (LRU), First-In-First-Out (FIFO),

Most-recently-used (MRU) etc.

• Policy can have big impact on # of I/O’s; depends on the access pattern.

• Sequential flooding: Nasty situation caused by LRU + repeated sequential scans.– # buffer frames < # pages in file means each page

request causes an I/O. MRU much better in this situation (but not in all situations, of course).

Access Manager• Arrange records into a

page• Retrieve records from a

page• What to concern

– Mapping between record and page

– Record formats– Page formats– File formats

Record 1Record 2

:::::Record n1

Page 1Page 2

:::::Page m

Insert/delete/retrieve a record from/to a page

Db1:T1Record 1Record 2

Record n2

Dbi:Tj

:: :::::

Record-Page Mapping • Maintain a mapping table

Record 1Record 2

:::::Record n1

Page 1Page 2

:::::Page m

Insert/delete/retrieve a record from/to a page

Db1:T1Record 1Record 2

Record n2

Dbi:Tj

:: :::::

DBdb1db2

dbn:::

t1t2

tm:::

Tablep1p2

pj:::

Page

t1t2

tm:::

Table R1R2R3

. . N

Record Formats: Fixed Length

• Information about field types and lengths is stored in system catalogs.• Li: Size of field i in bytes

Base address (B)

L1 L2 L3 L4

F1 F2 F3 F4

Address = B+L1+L2

Name: char (40) Address: char (100)Phone: char (10)Email: char (100)

Record Formats: Variable Length• Two alternative formats (# fields is fixed):

Second offers direct access to i’th field, efficient storage of nulls; small directory overhead.

4 $ $ $ $

FieldCount

Fields Delimited by Special Symbols

F1 F2 F3 F4

F1 F2 F3 F4

Array of Field Offsets

Can be used for fixed length fieldsOr Variable length fields

VARCHARBLOB

Page Formats: Fixed Length Records

Slot 1Slot 2

Slot N. . .

N

Solution 1: PACKED

FreeSpace

number of records

Each slot holds one record Record the number of records

(total = PageSize/RecordSize) The free slot starts from N+1 When appending a new

record, allocate slot N+1, then update N++

When deleting a record at slot i, move the last record to the slot i. If sorted, all records after slot i must be moved up

Slot 3

Page Formats: Fixed Length Records

. . .

M10. . .M ... 3 2 1

Slot 1Slot 2

Slot N

FreeSpace

Slot M11

numberof slots

Each slot holds one record Total number of slots is

PageSize/RecordSize Need a bitmap to record if a

slot is occupied or not If bit[i]==1, slot i is occupied When inserting a record,

search the bitmap to find a bit that is 0, then allocate the corresponding slot for the record

When deleting a record, simply reset the corresponding bit

Solution 1: UNPACKED

Page Formats: Variable Length Records

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

Page iRid = (i,N)

Rid = (i,2)

Rid = (i,1)

Pointerto startof freespace

SLOT DIRECTORY

N . . . 2 120 16 24 N

# slots

• Format• Heap File: Suitable when typical access is a file scan

retrieving all records.• Sorted File: Best if records must be retrieved in some

order, or only a `range’ of records is needed.• Hashed File: Good for equality selections.

File Formats and Operation Costs

• Cost factors (we ignore CPU costs, for simplicity)– P: The number of data pages– R: Number of records per page– D: (Average) time to read or write disk page

– Measuring number of page I/O’s ignores gains of pre-fetching blocks of pages; thus, even I/O cost is only approximated.

• The data in a heap file is not ordered. – How to find a page that has some free space– How to find the free space inside a page

– Depend on record format, i.e., fixed-length or variable length

• Two types of implementations– Link-based– Directory-based

Heap Files

Heap File Implemented as a List

• To insert a record, one searches the pages with free space and find the one that has sufficient space• Many pages may contain some tiny free space• A long list of pages may have to loaded in order to find an

appropriate one

HeaderPage

DataPage

DataPage

DataPage

DataPage

DataPage

DataPage Pages with

Free Space

Full Pages

Heap File Using a Page Directory

• The directory is a collection of pages; • Each page contains a number of entry• Each entry contains a pointer linking to the page and

a variable recording the free space• The number of directory pages is much smaller than

that of data pages

DataPage 1

DataPage 2

DataPage N

HeaderPage

DIRECTORY

Scan: P*DSearch with equality selection: If a selection is based on a candidate key, on average, we must scan half the file, assuming that the record exists 0.5*P*D.

Search with range selection: The entire file must be scanned. The cost is P*D.

Insert: Assume that records are always inserted at the end of the file. We fetch the last page in the file, add the record, and write the page back. The cost is 2D.

Delete:. The cost also depends on the number of qualifying records. The cost is search cost + D.

Heap Files and Associated Costs

P: The number of data pagesR: Number of records per pageD: (Average) time to read or write disk page

DataPage 1

DataPage 2

DataPage P

::directory

Sorted File

DataPage 2

DataPage N

HeaderPage

DIRECTORY

sorted field

DataPage 1

1

234

RID

Sort records directly• Expensive when inserting a record

Sorted File

DataPage n

Index Page

DIRECTORY

sorted field

DataPage 1

1

234

RID

• Keep the sorted field using index page• Each entry points to a record

• May contain the value of the sorted field• May contain a valid bit for deleting operation

• The entries are sorted according to the sorted field• Inserting a record just need to reorganize the index page

• The size is much smaller

Sorted Files and Associated CostsScan: P*DSearch with equality selection: Assume that the selection is specified on the field by which the file is sorted. The cost is D * log P assuming that the sorted file is stored sequentially.

Search with range selection:

Insert: Search cost + 2*0.5*P*D; the assumption is that the inserted record belongs in the middle of the file.

Delete: Search cost + 2*0.5*P*D; the assumption is that we need to pack the file and the record to be deleted is in the middle of the file.

D * log P + cost of retrieving qualified records.

Smith, 40, 3000Jones, 40, 6003Tracy, 40, 5004

h1ageh(age)=00

h(age)=01

h(age)=9

File hashed on age

•File is a collection of buckets. •Bucket = primary page plus zero or more overflow pages.

•Hashing function h: h(r) = bucket in which record r belongs. h looks at only some of the fields of r, called the search fields.

Hashed files

Doug, 20, 3800

Ashby, 21,3000Basu, 31, 4003Bristow, 21, 2007

Class, 59, 5004Daniels, 29, 6003

overflow pages

Smith, 40, 3000Jones, 40, 6003Tracy, 40, 5004

h1ageh(age)=00

h(age)=01

h(age)=9

File hashed on age

•File is a collection of buckets. •Bucket = primary page plus zero or more overflow pages.

•Hashing function h: h(r) = bucket in which record r belongs. h looks at only some of the fields of r, called the search fields.

Hashed files

Doug, 20, 3800

Ashby, 21,3000Basu, 31, 4003Bristow, 21, 2007

Class, 59, 5004Daniels, 29, 6003

overflow pages

Efficiency depends on• Hash function• Data skew factor

Hashed Files and Associated CostsAssume that there is no overflow page.

Scan: 1.25*P*D if pages are kept at 80% occupancy

Search with equality selection:D

Search with range selection: 1.25*P*D

Insert: Search cost + D = 2D

Delete: Search cost + D = 2D

HeapFile

Sorted File

HashedFile

Scan all records BD BD 1.25 BDEquality Search 0.5 BD D log 2B DRange Search BD D (log 2B + # of

pages withmatches)

1.25 BD

Insert 2D Search + BD 2DDelete Search + D Search + BD 2D

Several assumptions underlie these (rough) estimates!

Cost Comparison