DIsk BFR

CAS CS 460/660
Introduction to Database Systems
Disks, Buffer Manager
1.1
DBMS architecture
Database app
Query Optimization
and Execution
Relational Operators
These layers
Access Methods must consider
concurrency
Buffer Management control and
recovery
Disk Space Management
Student Records
stored on disk
1.2
Why Not Store It All in Main Memory?
 Costs too much. $100 will buy you either

~100 GB of RAM or around 2000 GB (2 TB) of disk
today.
 High-end Databases today can be in the Petabyte (1000TB)
range.
 Approx 60% of the cost of a production system is in the disks.
 Main memory is volatile. We want data to be saved
between runs. (Obviously!)
 Note, some specialized systems do store entire
database in main memory.
 Vendors claim 10x speed up vs. traditional DBMS running in
main memory.
1.3
The Storage Hierarchy
• Main memory (RAM) for

currently used data.
Small, Fast
• Disk for main database Registers
(secondary storage).
On-chip Cache
• Tapes for archive, maybe
(tertiary storage). On-Board Cache
• The role of Flash (SSD)

still unclear RAM
SSD
Disk Big, Slow

Tape
1.4
Economics
For $1000, you can get:

 ~400GB of RAM
 ~2.5TB of Solid State Disk
 ~30TB of Magnetic Disk
1.5
Disks and Files
 Today: Most data is stored on magnetic disks.

 Disks are a mechanical anachronism!
 Major implications!
 No “pointer derefs”. Instead, an API:
 READ: transfer “page” of data from disk to RAM.
 WRITE: transfer “page” of data from RAM to disk.
 Both API calls expensive
 Plan carefully!
 An explicit API can be a good thing
 Minimizes the kind of pointer errors you see in C
1.6
1.7
Anatomy of a Disk
The platters spin

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 size is a multiple of
sector size (which is fixed)
1.8
1.9
Accessing a Disk Page
 Time to access (read/write) a disk 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 )
 Seek time and rotational delay dominate.
Wait
 Seek time varies from about 1 to 15msec (full stroke)
 Rotational delay varies from 0 to 8msec (7200rpm)
 Transfer rate is < 0.1msec per 8KB block
 Key to lower I/O cost:
reduce seek/rotation delays! Hardware vs.
Transfer Transfer
software solutions?
Rotate
 Also note: For shared disks most time spent waiting in
queue for access to arm/controller Rotate
Seek
Seek
1.10
Arranging Pages on Disk
 `Next’ block concept:

 blocks on same track, followed by
 blocks on same cylinder, followed by
 blocks on adjacent cylinder
 Arrange file pages sequentially on disk
 minimize seek and rotational delay.
 For a sequential scan, pre-fetch
 several pages at a time!
 We use Block/Page interchangeably!!
1.11
From the DB Administrator’s View
Modern disk structures are so complex even

industry experts refer to them as “black
boxes”. Today there is no alignment to
physical disk sectors, no matter what we
believe. Disks do not map sectors to physical
regions in a way that we can understand from
outside the box; the simplistic “geometry”
reported by the device is an artifice.
from Microsoft’s “Disk Partition Alignment Best

Practices for SQL Server”
1.12
Disk Space Management
 Lowest layer of DBMS software manages

space on disk (using OS file system or not?).
 Higher levels call upon this layer to:
 allocate/de-allocate a page
 read/write a page
 Best if a request for a sequence of pages is
satisfied by pages stored sequentially on
disk! Higher levels don’t need to know if/how
this is done, or how free space is managed.
1.13
Notes on Flash (SSD)
 Various technologies, we focus on NAND

 suited for volume data storage
 alternative: NOR Flash
 Read is random access and fast
 E.g. 512 Bytes at a time
 Write is coarser grained and slower
 E.g. 16-512 KBytes at a time.
 Can get slower over time
 Some concern about write endurance
 100K cycle lifetimes?
 Still changing quickly
1.14
CPU Typical
Computer
... ...
M C
Secondary
Storage
1.15
Storage Pragmatics & Trends
 Many significant DBs are not that big.

 Daily weather, round the globe, 1929-2009: 20GB
 2000 US Census: 200GB
 2009 English Wikipedia: 14GB
 NYC Taxi Rides (~20GB per year)
 But data sizes grow faster than Moore’s Law
 What is the role of disk, flash, RAM?
 The subject of much debate/concern!
1.16
Buffer Management in a DBMS
Page Requests from Higher Levels
BUFFER POOL
disk page
free frame
MAIN MEMORY
choice of frame dictated
DISK by replacement policy
DB
 Data must be in RAM for DBMS to operate on it!

 The query processor refers to data using virtual memory
addresses.
 Buffer Mgr hides the fact that not all data is in RAM
1.17
Some Terminology…
 Disk Page – the unit of transfer between the disk and
memory
Typically set as a config parameter for the DBMS.
Typical value between 4 KBytes to 32 KBytes.
 Frame – a unit of memory
Typically the same size as the Disk Page Size
 Buffer Pool – a collection of frames used by the
DBMS to temporarily keep data for use by the query
processor (CPU).
 note: We will sometime use the term “buffer” and “frame”
synonymously.
1.18
When a Page is Requested ...
 If requested page IS in the pool:
 Pin the page and return its address.
 Else, if requested page IS NOT in the pool:
 If a free frame exists, choose it, Else:
 Choose a frame for replacement (only un-pinned

pages are candidates)
 If chosen frame is “dirty”, write it to disk
 Read requested page into chosen frame
 Pin the page and return its address.
Q: What information about buffers and their

contents must the system maintain?
1.19
Buffer Control Blocks (BCBs):
<frame#, pageid, pin_count, dirty>
 A page may be requested many times, so
• a pin count is used.
• To pin a page, pin_count++
• A page is a candidate for replacement iff pin_count
== 0 (“unpinned”)
 Requestor of page must eventually unpin it.
• pin_count--
 Must also indicate if page has been modified:
• dirty bit is used for this.
Q: Why is this important?
Q: How should BCB’s be organized?
1.20
Additional Buffer Mgr Notes
 BCB’s are hash indexed by pageID
 Concurrency Control & Recovery may entail additional I/O when a

frame is chosen for replacement.
(Write-Ahead Log protocol; more later.)
 If requests can be predicted (e.g., sequential scans) pages can be

pre-fetched several pages at a time.
1.21
Buffer Replacement Policy
 Frame is chosen for replacement by a
replacement policy:
 Least-recently-used (LRU), MRU, Clock, etc.
 This policy can have big impact on the number

of disk reads and writes.
 Remember, these are slooooooooooow.
 BIG IDEA – throw out the page that you are

least likely to need in the future.
 Q: How do you predict the future?
1.22
LRU Replacement Policy
Least Recently Used (LRU)

1)for each page in buffer pool, keep track of time
last unpinned
2)Replace the frame that has the oldest (earliest)
time
 Most common policy: intuitive and simple
 Based on notion of “Temporal Locality”
 Works well to keep “working set” in buffers.
 Implemented through doubly linked list of BCBs

 Requires list manipulation on unpin
1.23
“Clock” Replacement Policy
A(1)
 An approximation of LRU
D(1) B(p)
 Arrange frames into a cycle, store one
reference bit per frame
C(1)
 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;
else if (pincount == 0 && ref bit is off)
choose this page for
replacement;
} until a page is chosen;
1.24
Some issues with LRU
 Problem: Sequential flooding

 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).
 Problem: “cold” pages can hang around a long
time before they are replaced.
1.25
DBMS vs. OS File System
OS does disk space & buffer mgmt: why not let OS manage these tasks?
 Some limitations, e.g., files can’t span disks.

 Note, this is changing --- OS File systems are getting
smarter (i.e., more like databases!)
 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.
1.26

DIsk BFR

Uploaded by

Copyright:

Available Formats

DIsk BFR

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DIsk BFR

Uploaded by

Copyright:

Available Formats

CAS CS 460/660

Introduction to Database Systems

Disks, Buffer Manager

 Costs too much. $100 will buy you either

• Main memory (RAM) for

• The role of Flash (SSD)

Disk Big, Slow

For $1000, you can get:

 Today: Most data is stored on magnetic disks.

The platters spin

 `Next’ block concept:

 We use Block/Page interchangeably!!

Modern disk structures are so complex even

from Microsoft’s “Disk Partition Alignment Best

 Lowest layer of DBMS software manages

 Various technologies, we focus on NAND

 Many significant DBs are not that big.

 Data must be in RAM for DBMS to operate on it!

 Choose a frame for replacement (only un-pinned

Q: What information about buffers and their

 BCB’s are hash indexed by pageID

 Concurrency Control & Recovery may entail additional I/O when a

 If requests can be predicted (e.g., sequential scans) pages can be

 This policy can have big impact on the number

 BIG IDEA – throw out the page that you are

Least Recently Used (LRU)

 Implemented through doubly linked list of BCBs

 Problem: Sequential flooding

 Some limitations, e.g., files can’t span disks.

 Buffer management in DBMS requires ability to:

You might also like