db2 01

Storage
Torsten Grust
Chapter 2
Storage
Disks, Buffer Manager, Files. . . Magnetic Disks
Access Time
Sequential vs. Random
Architecture and Implementation of Database Systems Access
I/O Parallelism
Summer 2014 RAID Levels 1, 0, and 5
Alternative Storage
Techniques
Solid-State Disks
Network-Based Storage
Managing Space
Free Space Management
Buffer Manager
Pinning and Unpinning
Replacement Policies
Databases vs.
Operating Systems
Files and Records

Heap Files
Torsten Grust Inside a Page
Alternative Page Layouts
Wilhelm-Schickard-Institut für Informatik
Recap
Universität Tübingen 1
Storage
Database Architecture
Torsten Grust
Web Forms Applications SQL Interface
SQL Commands
Magnetic Disks
Access Time
Executor Parser Sequential vs. Random
Access
I/O Parallelism
Operator Evaluator Optimizer
RAID Levels 1, 0, and 5
Alternative Storage
Techniques
Transaction Files and Access Methods Solid-State Disks
Manager
Recovery Managing Space
Buffer Manager
Manager Free Space Management
Lock Manager Buffer Manager

Disk Space Manager Pinning and Unpinning
Databases vs.
DBMS Operating Systems
Files and Records

data files, indices, . . . Database Heap Files
Inside a Page
Recap
2
Storage
The Memory Hierarchy
Torsten Grust
capacity latency
CPU
(with registers) bytes < 1 ns
Magnetic Disks
Access Time
caches kilo-/megabytes < 10 ns Sequential vs. Random
Access
I/O Parallelism
main memory gigabytes 70–100 ns RAID Levels 1, 0, and 5
Alternative Storage
hard disks terabytes 3–10 ms Techniques
Solid-State Disks
tape library petabytes varies Managing Space
Buffer Manager
• Fast—but expensive and small—memory close to CPU Pinning and Unpinning

• Larger, slower memory at the periphery Databases vs.

Operating Systems
• DBMSs try to hide latency by using the fast memory as a Files and Records
cache. Heap Files

Inside a Page
Recap
3
Storage
Magnetic Disks
Torsten Grust
rotation
arm
track
Magnetic Disks
Access Time
Access
I/O Parallelism
Alternative Storage
block sector Techniques
platter heads Solid-State Disks

Managing Space
Photo: http://www.metallurgy.utah.edu/
A stepper motor positions an array of Buffer Manager
disk heads on the requested track Pinning and Unpinning

• Platters (disks) steadily rotate Databases vs.

Operating Systems
• Disks are managed in blocks: the system Files and Records
reads/writes data one block at a time Heap Files

Inside a Page
Recap
4
Storage
Access Time
Torsten Grust
Data blocks can only be read and written if disk heads

and platters are positioned accordingly.
Magnetic Disks
• This design has implications on the access time to

Access Time
Access
read/write a given block: I/O Parallelism
Definition (Access Time) Alternative Storage
Techniques
1 Move disk arms to desired track (seek time ts ) Solid-State Disks

2 Disk controller waits for desired block to rotate under Managing Space
disk head (rotational delay tr ) Free Space Management
Buffer Manager
3 Read/write data (transfer time ttr ) Pinning and Unpinning
Databases vs.
⇒ access time: t = ts + tr + ttr Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
5
Storage
Example: Seagate Cheetah 15K.7
Torsten Grust
(600 GB, server-class drive)
• Seagate Cheetah 15K.7 performance characteristics:

• 4 disks, 8 heads, avg. 512 kB/track, 600 GB capacity
• rotational speed: 15 000 rpm (revolutions per minute) Magnetic Disks
Access Time
• average seek time: 3.4 ms Sequential vs. Random
Access
• transfer rate ≈ 163 MB/s I/O Parallelism
What is the access time to read an 8 KB data block? Alternative Storage

Techniques
Solid-State Disks
Managing Space
Buffer Manager
Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
6
Storage
Example: Seagate Cheetah 15K.7
Torsten Grust
(600 GB, server-class drive)
• Seagate Cheetah 15K.7 performance characteristics:

• 4 disks, 8 heads, avg. 512 kB/track, 600 GB capacity
• rotational speed: 15 000 rpm (revolutions per minute) Magnetic Disks
Access Time
• average seek time: 3.4 ms Sequential vs. Random
Access

Techniques
Solid-State Disks
average seek time ts = 3.40 ms Network-Based Storage
1 1 Managing Space
average rotational delay: 2 · 15 000 min−1 tr = 2.00 ms Free Space Management
8 kB Buffer Manager
transfer time for 8 KB: 163 MB/s ttr = 0.05 ms Pinning and Unpinning
Databases vs.
Operating Systems
access time for an 8 kB data block t = 5.45 ms
Files and Records
Heap Files
Inside a Page
Recap
6
Storage
Sequential vs. Random Access
Torsten Grust
Example (Read 1 000 blocks of size 8 kB)
• random access:
Magnetic Disks
trnd = 1 000 · 5.45 ms = 5.45 s Access Time
• sequential read of adjacent blocks:

Access
I/O Parallelism
tseq = ts + tr + 1 000 · ttr + 16 · ts,track-to-track RAID Levels 1, 0, and 5
= 3.40 ms + 2.00 ms + 50 ms + 3.2 ms ≈ 58.6 ms Alternative Storage

Techniques
The Seagate Cheetah 15K.7 stores an average of 512 kB per Solid-State Disks
track, with a 0.2 ms track-to-track seek time; our 8 kB blocks
Managing Space
are spread across 16 tracks. Free Space Management
Buffer Manager
⇒ Sequential I/O is much faster than random I/O Replacement Policies
Databases vs.

⇒ Avoid random I/O whenever possible Operating Systems
Files and Records

58.6 ms
⇒ As soon as we need at least 5,450 ms = 1.07 % of a file,
Heap Files
we better read the entire file sequentially Inside a Page
Recap
7
Storage
Performance Tricks
Torsten Grust
• Disk manufacturers play a number of tricks to improve

performance:
track skewing Magnetic Disks
Align sector 0 of each track to avoid Access Time

rotational delay during longer Access
I/O Parallelism
sequential scans RAID Levels 1, 0, and 5
Alternative Storage
Techniques
request scheduling Solid-State Disks

If multiple requests have to be served, choose the one Managing Space
that requires the smallest arm movement (SPTF: shortest Free Space Management
Buffer Manager
positioning time first, elevator algorithms) Pinning and Unpinning
Databases vs.
zoning Operating Systems
Files and Records

Outer tracks are longer than the inner ones. Therefore, Heap Files
divide outer tracks into more sectors than inner tracks Free Space Management
Inside a Page
Recap
8
Storage
Evolution of Hard Disk Technology
Torsten Grust
Disk seek and rotational latencies have only marginally improved

over the last years (≈ 10 % per year)
Magnetic Disks
But: Access Time

Access
• Throughput (i.e., transfer rates) improve by ≈ 50 % per year I/O Parallelism
• Hard disk capacity grows by ≈ 50 % every year

Alternative Storage
Techniques
Solid-State Disks
Therefore: Network-Based Storage
Managing Space
• Random access cost hurts even more as time progresses Free Space Management
Buffer Manager
Example (5 Years Ago: Seagate Barracuda 7200.7) Databases vs.

Operating Systems
Read 1K blocks of 8 kB sequentially/randomly: 397 ms / 12 800 ms Files and Records
Heap Files
Inside a Page
Recap
9
Storage
Ways to Improve I/O Performance
Torsten Grust
The latency penalty is hard to avoid
But:
• Throughput can be increased rather easily by exploiting Magnetic Disks
Access Time
parallelism Sequential vs. Random

Access
• Idea: Use multiple disks and access them in parallel, try to I/O Parallelism
hide latency Alternative Storage
Techniques
Solid-State Disks
TPC-C: An industry benchmark for OLTP Network-Based Storage
Managing Space
A recent #1 system (IBM DB2 9.5 on AIX) uses Free Space Management
Buffer Manager
• 10,992 disk drives (73.4 GB each, 15,000 rpm) (!) Pinning and Unpinning
plus 8 146.8 GB internal SCSI drives, Databases vs.
• connected with 68 4 Gbit fibre channel adapters, Operating Systems
Files and Records

• yielding 6 mio transactions per minute Heap Files
Inside a Page
Recap
10
Storage
Disk Mirroring
Torsten Grust
• Replicate data onto multiple disks:
Magnetic Disks
1 2 3 4 5 Access Time
6 7 8 9 ··· Sequential vs. Random

Access
1 2 3 4 5 I/O Parallelism
6 7 8 9 ···
Alternative Storage
Techniques
1 2 3 4 5
Solid-State Disks
6 7 8 9 ··· Network-Based Storage
Managing Space
Buffer Manager
• Achieves I/O parallelism only for reads Replacement Policies
Databases vs.
• Improved failure tolerance—can survive one disk failure Operating Systems
• This is also known as RAID 1 (mirroring without parity)

Files and Records
Heap Files
(RAID: Redundant Array of Inexpensive Disks) Free Space Management

Inside a Page
Recap
11
Storage
Disk Striping
Torsten Grust
• Distribute data over disks:

Magnetic Disks
Access Time
1 2 3 4 5 Sequential vs. Random
Access
6 7 8 9 ···
I/O Parallelism
Alternative Storage
Techniques
1 4 7 ··· 2 5 8 ··· 3 6 9 ··· Solid-State Disks
Managing Space
Buffer Manager
• Full I/O parallelism for read and write operations

• High failure risk (here: 3 times risk of single disk failure)! Databases vs.
Operating Systems
• Also known as RAID 0 (striping without parity) Files and Records

Heap Files
Inside a Page
Recap
12
Storage
Disk Striping with Parity
Torsten Grust
• Distribute data and parity information over > 3 disks:
1 2 3 4 5 Magnetic Disks
6 7 8 9 ··· Access Time
Access
I/O Parallelism
1 3 5/6 7 · · · 2 3/4 5 8 · · · 1/2 4 6 7/8 · · · Alternative Storage

Techniques
Solid-State Disks
Managing Space
• High I/O parallelism Buffer Manager
• Fault tolerance: any one disk may fail without data loss
(with dual parity/RAID 6: two disks may fail) Databases vs.

Operating Systems
• Distribute parity (e.g., XOR) information over disks, separating Files and Records
data and associated parity Heap Files

• Also known as RAID 5 (striping with distributed parity) Inside a Page

Recap
13
Storage
Solid-State Disks
Torsten Grust
Solid state disks (SSDs) have emerged as an alternative to

conventional hard disks
• SSDs provide very low-latency

Magnetic Disks
random read access (< 0.01 ms) Access Time
• Random writes, however, are

write
Access
time
significantly slower than on I/O Parallelism
read
write
traditional magnetic drives:
read
Alternative Storage
Techniques
1 (Blocks of ) Pages have to be flash mag. disk
Solid-State Disks
erased before they can be
Managing Space
updated Free Space Management
2 Once pages have been Buffer Manager

erased, sequentially writing Replacement Policies
them is almost as fast as Databases vs.

Operating Systems
reading
Files and Records
Heap Files
Inside a Page
Recap
14
Storage
SSDs: Page-Level Writes, Block-Level Deletes
Torsten Grust
• Typical page size: 128 kB

• SSDs erase blocks of pages: block ≈ 64 pages (8 MB)
Example (Perform block-level delete to accomodate new data pages)

Magnetic Disks
Access Time
Access
I/O Parallelism
Alternative Storage
Techniques
Solid-State Disks
Illustration taken from arstechnica.com (The SSD Revolution)
Managing Space
Buffer Manager
Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
15
Storage
Example: Seagate Pulsar.2
Torsten Grust
(800 GB, server-class solid state drive)
• Seagate Pulsar.2 performance characteristics:

• NAND flash memory, 800 GB capacity
• standard 2.500 enclosure, no moving/rotating parts Magnetic Disks
Access Time
• data read/written in pages of 128 kB size Sequential vs. Random
Access

Techniques
Solid-State Disks
Managing Space
Buffer Manager
Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
16
Storage
Example: Seagate Pulsar.2
Torsten Grust
(800 GB, server-class solid state drive)
• Seagate Pulsar.2 performance characteristics:

• NAND flash memory, 800 GB capacity
• standard 2.500 enclosure, no moving/rotating parts Magnetic Disks
Access Time
• data read/written in pages of 128 kB size Sequential vs. Random
Access

Techniques
Solid-State Disks
no seek time ts = 0.00 ms Network-Based Storage
Managing Space
no rotational delay: tr = 0.00 ms Free Space Management
128 kB
transfer time for 8 KB: 370 MB/s ttr = 0.30 ms Buffer Manager
Databases vs.
Operating Systems
access time for an 8 kB data block t = 0.30 ms
Files and Records
Heap Files
Inside a Page
Recap
16
Storage
Sequential vs. Random Access with SSDs
Torsten Grust
Example (Read 1 000 blocks of size 8 kB)
• random access: Magnetic Disks

Access Time
trnd = 1 000 · 0.30 ms = 0.3 s Sequential vs. Random
Access
• sequential read of adjacent pages: I/O Parallelism

· 8 kB
tseq = 1 000
128 kB · ttr ≈ 18.9 ms Alternative Storage
Techniques
The Seagate Pulsar.2 (sequentially) reads data in 128 kB Solid-State Disks

chunks. Managing Space

Buffer Manager
⇒ Sequential I/O still beats random I/O Pinning and Unpinning
(but random I/O is more feasible again) Databases vs.
• Adapting database technology to these characteristics is a

Operating Systems
Files and Records

current research topic Heap Files
Inside a Page
Recap
17
Storage
Torsten Grust
Today the network is not a bottleneck any more:
Magnetic Disks
Storage media/interface Transfer rate Access Time
Hard disk 100–200 MB/s Access
Serial ATA 375 MB/s I/O Parallelism

Ultra-640 SCSI 640 MB/s Alternative Storage
10-Gbit Ethernet 1,250 MB/s Techniques
Solid-State Disks
Infiniband QDR 12,000 MB/s Network-Based Storage
Managing Space
For comparison (RAM):
Buffer Manager
PC2–5300 DDR2–SDRAM 10.6 GB/s Pinning and Unpinning
PC3–12800 DDR3–SDRAM 25.6 GB/s Replacement Policies
Databases vs.
Operating Systems
Files and Records

⇒ Why not use the network for database storage? Heap Files
Inside a Page
Recap
18
Storage
Storage Area Network (SAN)
Torsten Grust
• Block-based network access to storage:

• SAN emulate interface of block-structured disks Magnetic Disks
(“read block #42 of disk 10”) Access Time
• This is unlike network file systems (e.g., NFS, CIFS)

Access
I/O Parallelism
• SAN storage devices typically abstract from RAID or physical Alternative Storage
Techniques
disks and present logical drives to the DBMS Solid-State Disks
• Hardware acceleration and simplified maintainability Managing Space

Buffer Manager
• Typical setup: local area network with multiple participating

servers and storage resources Databases vs.

Operating Systems
• Failure tolerance and increased flexibility Files and Records

Heap Files
Inside a Page
Recap
19
Storage
Grid or Cloud Storage
Torsten Grust
Internet-scale enterprises employ clusters with 1000s commodity

PCs (e.g., Amazon, Google, Yahoo!):
• system cost ↔ reliability and performance,
Magnetic Disks
• use massive replication for data storage Access Time
Spare CPU cycles and disk space are sold as a service: Access
I/O Parallelism
Amazon’s Elastic Computing Cloud (EC2) RAID Levels 1, 0, and 5
Alternative Storage
Use Linux-based compute cluster by the hour (∼ 10 ¢/h). Techniques
Solid-State Disks
Amazon’s Simple Storage System (S3) Network-Based Storage
“Infinite” store for objects between 1 B and 5 TB in size, Managing Space

organized in a map data structure (key 7→ object) Buffer Manager
• Latency: 100 ms to 1 s (not impacted by load) Pinning and Unpinning
• pricing ≈ disk drives (but addl. cost for access) Databases vs.
Operating Systems
⇒ Building a database on S3? Files and Records

Heap Files
(% Brantner et al., SIGMOD 2008) Free Space Management
Inside a Page
Recap
20
Storage
Managing Space
Torsten Grust
SQL Commands
Magnetic Disks
Executor Parser Access Time
Access
I/O Parallelism
Alternative Storage
Transaction Files and Access Methods Techniques
Solid-State Disks
Manager
Recovery Network-Based Storage
Buffer Manager
Manager Managing Space
Lock Manager
Disk Space Manager Buffer Manager
DBMS Databases vs.
Operating Systems
data files, indices, . . . Database Files and Records

Heap Files
Inside a Page
Recap
21
Storage
Managing Space
Torsten Grust
Definition (Disk Space Manager)
• Abstracts from the gory details of the underlying storage

(disk space manager talks to I/O controller and initiates I/O) Magnetic Disks
Access Time
• DBMS issues allocate/deallocate and read/write Sequential vs. Random

Access
commands I/O Parallelism
• Provides the concept of a page (typically 4–64 KB) as a unit

Alternative Storage
of storage to the remaining system components Techniques
Solid-State Disks
• Maintains a locality-preserving mapping Network-Based Storage
Managing Space
page number 7→ physical location , Buffer Manager
where a physical location could be, e.g., Replacement Policies
• an OS file name and an offset within that file, Databases vs.

Operating Systems
• head, sector, and track of a hard drive, or Files and Records
• tape number and offset for data stored in a tape library Heap Files
Inside a Page
Recap
22
Storage
Empty Pages
Torsten Grust
The disk space manager also keeps track of used/free blocks

(deallocation and subsequent allocation may create holes):
1 Maintain a linked list of free pages
• When a page is no longer needed, add it to the list Magnetic Disks
Access Time
2 Maintain a bitmap reserving one bit for each page Access
• Toggle bit n when page n is (de-)allocated I/O Parallelism

Alternative Storage
Allocation of contiguous pages

Techniques
Solid-State Disks
Managing Space
To exploit sequential access, it is useful to allocate contiguous Free Space Management
sequences of pages. Buffer Manager

Which of the techniques (1 or 2) would you choose to support Replacement Policies
this? Databases vs.

Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
23
Storage
Empty Pages
Torsten Grust
The disk space manager also keeps track of used/free blocks

(deallocation and subsequent allocation may create holes):
1 Maintain a linked list of free pages
• When a page is no longer needed, add it to the list Magnetic Disks
Access Time
2 Maintain a bitmap reserving one bit for each page Access
• Toggle bit n when page n is (de-)allocated I/O Parallelism

Alternative Storage
Allocation of contiguous pages

Techniques
Solid-State Disks
Managing Space
To exploit sequential access, it is useful to allocate contiguous Free Space Management
sequences of pages. Buffer Manager

Which of the techniques (1 or 2) would you choose to support Replacement Policies
this? Databases vs.

Operating Systems
Files and Records

This is a lot easier to do with a free page bitmap (option 2). Heap Files
Inside a Page
Recap
23
Storage
Buffer Manager
Torsten Grust
SQL Commands
Magnetic Disks
Access
I/O Parallelism
Alternative Storage
Solid-State Disks
Manager
Buffer Manager
Lock Manager
DBMS Databases vs.
Operating Systems

Heap Files
Inside a Page
Recap
24
Storage
Buffer Manager
Torsten Grust
page requests Definition (Buffer Manager)

main
memory • Mediates between external Magnetic Disks
storage and main memory, Access Time
6 1 4 Sequential vs. Random
3
• Manages a designated main
Access
I/O Parallelism
memory area, the buffer RAID Levels 1, 0, and 5
7 pool, for this task Alternative Storage

Techniques
Solid-State Disks
disk page free frame Disk pages are brought into Managing Space
memory as needed and loaded Free Space Management
Buffer Manager
into memory frames Pinning and Unpinning
1 2 3 4 5 6 Replacement Policies
7 8 9 10 11 · · · A replacement policy decides Databases vs.

Operating Systems
which page to evict when the Files and Records

disk buffer is full Heap Files
Inside a Page
Recap
25
Storage
Interface to the Buffer Manager
Torsten Grust
Higher-level code requests (pins) pages from the buffer manager

and releases (unpins) pages after use.
pin (pageno)
Magnetic Disks
Request page number pageno from the buffer manager, load Access Time
it into memory if necessary and mark page as clean (¬dirty). Sequential vs. Random
Access
Returns a reference to the frame containing pageno. I/O Parallelism

unpin (pageno, dirty) Alternative Storage

Techniques
Release page number pageno, making it a candidate for Solid-State Disks
eviction. Must set dirty = true if page was modified. Network-Based Storage
Managing Space
Why do we need the dirty bit? Buffer Manager

Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
26
Storage
Interface to the Buffer Manager
Torsten Grust
Higher-level code requests (pins) pages from the buffer manager

and releases (unpins) pages after use.
pin (pageno)
Magnetic Disks
Request page number pageno from the buffer manager, load Access Time
it into memory if necessary and mark page as clean (¬dirty). Sequential vs. Random
Access
Returns a reference to the frame containing pageno. I/O Parallelism

unpin (pageno, dirty) Alternative Storage

Techniques
Release page number pageno, making it a candidate for Solid-State Disks
eviction. Must set dirty = true if page was modified. Network-Based Storage
Managing Space
Why do we need the dirty bit? Buffer Manager

Only modified pages need to be written back to disk upon Databases vs.
Operating Systems
eviction. Files and Records
Heap Files
Inside a Page
Recap
26
Storage
Proper pin ()/unpin () Nesting
Torsten Grust
• Any database transaction is required to properly “bracket”

any page operation using pin () and unpin () calls: Magnetic Disks
Access Time
A read-only transaction Sequential vs. Random

Access
a ← pin (p); I/O Parallelism

 RAID Levels 1, 0, and 5
..
 Alternative Storage
. Techniques
read data on page at memory address a ; Solid-State Disks


.. Managing Space

. Free Space Management
unpin (p, false); Buffer Manager

• Proper bracketing enables the systems to keep a count of Databases vs.

Operating Systems
active users (e.g., transactions) of a page
Files and Records
Heap Files
Inside a Page
Recap
27
Storage
Implementation of pin ()
Torsten Grust
Function pin(pageno)
1 if buffer pool already contains pageno then Magnetic Disks
2 pinCount (pageno) ← pinCount (pageno) + 1 ; Access Time

Access
3 return address of frame holding pageno ;
I/O Parallelism
4 else RAID Levels 1, 0, and 5
5 select a victim frame v using the replacement policy ; Alternative Storage

Techniques
6 if dirty (page in v) then Solid-State Disks
7 write page in v to disk ;
Managing Space
8 read page pageno from disk into frame v ; Free Space Management
Buffer Manager
9 pinCount (pageno) ← 1 ; Pinning and Unpinning
10 dirty (pageno) ← false ; Replacement Policies
11 return address of frame v ; Databases vs.

Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
28
Storage
Implementation of unpin ()
Torsten Grust
Function unpin(pageno, dirty)

1 pinCount (pageno) ← pinCount (pageno) − 1 ;
Magnetic Disks
2 dirty (pageno) ← dirty (pageno) ∨ dirty ; Access Time
Access
I/O Parallelism
Why don’t we write pages back to disk during unpin ()? RAID Levels 1, 0, and 5
Alternative Storage
Techniques
Solid-State Disks
Managing Space
Buffer Manager
Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
29
Storage
Implementation of unpin ()
Torsten Grust
Function unpin(pageno, dirty)

1 pinCount (pageno) ← pinCount (pageno) − 1 ;
Magnetic Disks
2 dirty (pageno) ← dirty (pageno) ∨ dirty ; Access Time
Access
I/O Parallelism
Why don’t we write pages back to disk during unpin ()? RAID Levels 1, 0, and 5
Alternative Storage
Techniques
Well, we could . . . Solid-State Disks

+ recovery from failure would be a lot simpler Managing Space
– higher I/O cost (every page write implies a write to disk) Free Space Management
Buffer Manager
– bad response time for writing transactions Pinning and Unpinning
This discussion is also known as force (or write-through) vs. Databases vs.
write-back. Actual database systems typically implement Operating Systems
write-back. Files and Records

Heap Files
Inside a Page
Recap
29
Storage
Concurrent Writes?
Torsten Grust
Conflicting changes to a block

Assume the following: Magnetic Disks
Access Time
1 The same page p is requested by more than one transaction Sequential vs. Random
Access
(i.e., pinCount (p) > 1), and I/O Parallelism

2 . . . those transactions perform conflicting writes on p? Alternative Storage
Techniques
Solid-State Disks
Conflicts of this kind are resolved by the system’s concurrency Network-Based Storage
Managing Space
control, a layer on top of the buffer manager (see “Introduction Free Space Management
to Database Systems”, transaction management, locks). Buffer Manager

The buffer manager may assume that everything is in order Databases vs.
whenever it receives an unpin (p, true) call. Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
30
Storage
Torsten Grust
The effectiveness of the buffer manager’s caching functionality

can depend on the replacement policy it uses, e.g.,
Least Recently Used (LRU)
Evict the page whose latest unpin () is longest ago
Magnetic Disks
LRU-k Access Time

Like LRU, but considers the k latest unpin () calls, not just Access
I/O Parallelism
the latest RAID Levels 1, 0, and 5
Alternative Storage
Most Recently Used (MRU) Techniques
Evict the page that has been unpinned most recently Solid-State Disks
Random Managing Space

Pick a victim randomly Buffer Manager
Rationales behind each of these policies?

Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
31
Storage
Example Policy: Clock (“Second Chance”)
Torsten Grust
• Simulate an LRU policy with less overhead (no LRU queue

reorganization on every frame reference):
Magnetic Disks
Access Time
Clock (“Second Chance”) Sequential vs. Random
Access
I/O Parallelism
1 Number the N frames in the buffer pool 0, . . . , N − 1; RAID Levels 1, 0, and 5
initialize current ← 0, maintain a bit array Alternative Storage

Techniques
referenced[0, . . . , N − 1] initialized to all 0 Solid-State Disks
2 In pin(p): load p into buffer pool (if needed), assign
Managing Space
referenced[frame-of(p)] ← 1 Free Space Management
Buffer Manager
3 In pin(p): if we need to find a victim, consider page Pinning and Unpinning
current; if referenced[current] = 0, current is the Replacement Policies
victim; otherwise, referenced[current] ← 0, Databases vs.

Operating Systems
current ← current + 1 mod N, repeat 3 Files and Records
Heap Files
Inside a Page
Recap
32
Storage
Heuristic Policies Can Fail
Torsten Grust
The mentioned policies, including LRU, are heuristics

only and may fail miserably in certain scenarios. Magnetic Disks
Access Time
Example (A Challenge for LRU) Access
I/O Parallelism
A number of transactions want to scan the same sequence of RAID Levels 1, 0, and 5
pages (consider a repeated SELECT * FROM R). Alternative Storage

Techniques
Assume a buffer pool capacity of 10 pages. Solid-State Disks
1 Let the size of relation R be 10 or less pages. Managing Space
How many I/O (actual disk page reads) do you expect? Free Space Management
Buffer Manager
2 Now grow R by one page. Pinning and Unpinning
How about the number of I/O operations in this case? Replacement Policies
Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
33
Storage
More Challenges for LRU
Torsten Grust
1 Transaction T1 repeatedly accesses a fixed set of pages;

transaction T2 performs a sequential scan of the database
pages. Magnetic Disks
2 Assume a B+-tree-indexed relation R. R occupies 10,000 data Access Time

pages Ri , the B+-tree occupies one root node and 100 index Access
I/O Parallelism
leaf nodes Ik . RAID Levels 1, 0, and 5
Transactions perform repeated random index key lookups on Alternative Storage
R ⇒ page access pattern (ignores B+-tree root node):

Techniques
Solid-State Disks
Managing Space
I1 , R1 , I2 , R2 , I3 , R3 , . . . Free Space Management
Buffer Manager
How will LRU perform in this case?

Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
34
Storage
More Challenges for LRU
Torsten Grust
1 Transaction T1 repeatedly accesses a fixed set of pages;

transaction T2 performs a sequential scan of the database
pages. Magnetic Disks
2 Assume a B+-tree-indexed relation R. R occupies 10,000 data Access Time

pages Ri , the B+-tree occupies one root node and 100 index Access
I/O Parallelism
leaf nodes Ik . RAID Levels 1, 0, and 5
Transactions perform repeated random index key lookups on Alternative Storage
R ⇒ page access pattern (ignores B+-tree root node):

Techniques
Solid-State Disks
Managing Space
I1 , R1 , I2 , R2 , I3 , R3 , . . . Free Space Management
Buffer Manager
How will LRU perform in this case?

Databases vs.
Operating Systems
With LRU, 50 % of the buffered pages will be pages of R. However, Files and Records
the probability of re-accessing page Ri only is 1/10,000. Heap Files
Inside a Page
Recap
34
Storage
Buffer Management in Practice
Torsten Grust
Prefetching
Buffer managers try to anticipate page requests to overlap
CPU and I/O operations:
• Speculative prefetching: Assume sequential scan and Magnetic Disks
Access Time
automatically read ahead. Sequential vs. Random

Access
• Prefetch lists: Some database algorithms can instruct I/O Parallelism
the buffer manager with a list of pages to prefetch. RAID Levels 1, 0, and 5
Alternative Storage
Techniques
Solid-State Disks
Page fixing/hating Network-Based Storage
Higher-level code may request to fix a page if it may be Managing Space

useful in the near future (e.g., nested-loop join). Buffer Manager
Likewise, an operator that hates a page will not access it any Replacement Policies
time soon (e.g., table pages in a sequential scan). Databases vs.

Operating Systems
Files and Records
Partitioned buffer pools Heap Files

E.g., maintain separate pools for indexes and tables. Inside a Page
Recap
35
Storage
Databases vs. Operating Systems
Torsten Grust

Wait! Didn’t we just re-invent the operating system?
Magnetic Disks
Yes, Access Time
• disk space management and buffer management very much

Access
look like file management and virtual memory in OSs. I/O Parallelism
But, Alternative Storage

Techniques
• a DBMS may be much more aware of the access patterns of Solid-State Disks
certain operators (prefetching, page fixing/hating), Managing Space
• concurrency control often calls for a prescribed order of Free Space Management
Buffer Manager
write operations, Pinning and Unpinning
• technical reasons may make OS tools unsuitable for a

Databases vs.
database (e.g., file size limitation, platform independence). Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
36
Storage
Databases vs. Operating Systems
Torsten Grust
In fact, databases and operating systems sometimes interfere:

• Operating system and buffer manager
effectively buffer the same data twice.
DBMS
buffer Magnetic Disks
• Things get really bad if parts of the Access Time

DBMS buffer get swapped out to disk Access
by OS VM manager. OS I/O Parallelism

buffer RAID Levels 1, 0, and 5
Alternative Storage
Techniques
• Therefore, database systems try to Solid-State Disks
turn off certain OS features or Managing Space
services. disk Free Space Management
Buffer Manager
⇒ Raw disk access instead of OS files. Replacement Policies
Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
37
Storage
Files and Records
Torsten Grust
SQL Commands
Magnetic Disks
Access
I/O Parallelism
Alternative Storage
Solid-State Disks
Manager
Buffer Manager
Lock Manager
DBMS Databases vs.
Operating Systems

Heap Files
Inside a Page
Recap
38
Storage
Database Files
Torsten Grust
• So far we have talked about pages. Their management is

oblivious with respect to their actual content.
• On the conceptual level, a DBMS primarily manages tables
of tuples and indexes.
• Such tables are implemented as files of records:
Magnetic Disks
Access Time
• A file consists of one or more pages, Sequential vs. Random

Access
• each page contains one or more records, I/O Parallelism
• each record corresponds to one tuple: RAID Levels 1, 0, and 5
Alternative Storage
Techniques
Solid-State Disks
Managing Space
file 0
Buffer Manager
free Replacement Policies
Databases vs.
page 0 page 1 page 2 page 3 Operating Systems
file 1 Files and Records
Heap Files
free Free Space Management

Inside a Page
page 4 page 5 page 6 page 7
Recap
39
Database Heap Files tumpukan file Storage
Torsten Grust
The most important type of files in a database is the heap file. It
stores records in no particular order (in line with, e.g., the SQL
semantics): penyimpanan gk pake urutan jelas
Typical heap file interface
Magnetic Disks
• create/destroy heap file f named n:

Access Time
Access
f = createFile(n), deleteFile(n) I/O Parallelism
• insert record r and return its rid: RAID Levels 1, 0, and 5
Alternative Storage
rid = insertRecord(f ,r) Techniques
• delete a record with a given rid:

Solid-State Disks
deleteRecord(f ,rid) Managing Space

• get a record with a given rid: Buffer Manager
r = getRecord(f ,rid) Pinning and Unpinning

• initiate a sequential scan over the whole heap file: Databases vs.
Operating Systems
openScan(f ) Files and Records
Heap Files
N.B. Record ids (rid) are used like record addresses (or pointers). Inside a Page
The heap file structure maps a given rid to the page containing
Recap
the record. 40
Storage
Heap Files
Torsten Grust
(Doubly) Linked list of pages:

Header page allocated when createFile(n) is
Magnetic Disks
called—initially both page lists are empty: Access Time
Access
data data data pages w/ I/O Parallelism

··· RAID Levels 1, 0, and 5
page page page free space Alternative Storage
header Techniques
page Solid-State Disks
data data data
page page
···
page full pages Managing Space
Buffer Manager
+ easy to implement Replacement Policies
Databases vs.
– most pages will end up in free page list Operating Systems
Files and Records

– might have to search many pages to place a (large) record Heap Files
Inside a Page
Recap
41
Storage
Heap Files
Torsten Grust
Operation insertRecord(f ,r) for linked list of pages
1 Try to find a page p in the free list with free space > | r |;
should this fail, ask the disk space manager to allocate a new
page p Magnetic Disks
Access Time
2 Write record r to page p Sequential vs. Random
Access
3 Since, generally, | r || p |, p will belong to the list of pages I/O Parallelism
with free space RAID Levels 1, 0, and 5
Alternative Storage
4 A unique rid for r is generated and returned to the caller Techniques
Solid-State Disks
Generating sensible record ids (rid) Managing Space

Buffer Manager
Given that rids are used like record addresses: what would be a Pinning and Unpinning
feasible rid generation method? Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
42
Storage
Heap Files
Torsten Grust
Operation insertRecord(f ,r) for linked list of pages
1 Try to find a page p in the free list with free space > | r |;
should this fail, ask the disk space manager to allocate a new
page p Magnetic Disks
Access Time
2 Write record r to page p Sequential vs. Random
Access
3 Since, generally, | r || p |, p will belong to the list of pages I/O Parallelism
with free space RAID Levels 1, 0, and 5
Alternative Storage
4 A unique rid for r is generated and returned to the caller Techniques
Solid-State Disks
Generating sensible record ids (rid) Managing Space

Buffer Manager
Given that rids are used like record addresses: what would be a Pinning and Unpinning
feasible rid generation method? Databases vs.
Generate a composite rid consisting of the address of page p and Operating Systems
the placement (offset/slot) of r inside p: Files and Records

Heap Files
hpageno p, slotno ri Inside a Page

Recap
42
Storage
Heap Files
Torsten Grust
Directory of pages:
Magnetic Disks
data Access Time
page data Sequential vs. Random

Access
page
I/O Parallelism
data RAID Levels 1, 0, and 5
page Alternative Storage

Techniques
···
Solid-State Disks
• Use as space map with information about free space on Managing Space
each page Buffer Manager
• granularity as trade-off space ↔ accuracy Pinning and Unpinning

(may range from open/closed bit to exact information) Databases vs.

Operating Systems
+ free space search more efficient Files and Records
Heap Files
– memory overhead to host the page directory Free Space Management
Inside a Page
Recap
43
Storage
Torsten Grust
Which page to pick for the insertion of a new record?

Append Only
Magnetic Disks
Always insert into last page. Otherwise, create a new page. Access Time
Best Fit Access
Reduces fragmentation, but requires searching the entire I/O Parallelism

free list/space map for each insert. Alternative Storage

Techniques
First Fit Solid-State Disks
Search from beginning, take first page with sufficient space. Managing Space
(⇒ These pages quickly fill up, system may waste a lot of Free Space Management
search effort in these first pages later on.) Buffer Manager

Next Fit Replacement Policies
Databases vs.
Maintain cursor and continue searching where search Operating Systems
stopped last time. Files and Records

Heap Files
Inside a Page
Recap
44
Storage
Free Space Witnesses
Torsten Grust
We can accelerate the search by remembering witnesses: Magnetic Disks

Access Time
• Classify pages into buckets, e.g., “75 %–100 % full”, Sequential vs. Random
Access
“50 %–75 % full”, “25 %–50 % full”, and “0 %–25 % full”. I/O Parallelism
• For each bucket, remember some witness pages. Alternative Storage
• Do a regular best/first/next fit search only if no witness is

Techniques
Solid-State Disks
recorded for the specific bucket.
Managing Space
• Populate witness information, e.g., as a side effect when Free Space Management
searching for a best/first/next fit page. Buffer Manager

Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
Recap
45
Storage
Inside a Page — Fixed-Length Records
Torsten Grust
Now turn to the internal page structure:
ID NAME SEX 4711John M1723M

arc M6381Betty Magnetic Disks
4711 John M F Access Time
1723 Marc M Sequential vs. Random

Access
6381 Betty F I/O Parallelism

Alternative Storage
• Record identifier (rid): Techniques

Solid-State Disks
hpageno, slotnoi Network-Based Storage
Managing Space
• Record position (within page): Free Space Management
slotno × bytes per slot Buffer Manager

no. of records
• Record deletion? in this page
Databases vs.
• record id should not Operating Systems
Files and Records

change Heap Files
⇒ slot directory (bitmap) 3 Header Free Space Management

Inside a Page
Recap
46
Storage
Inside a Page — Fixed-Length Records
Torsten Grust
Now turn to the internal page structure:
ID NAME SEX 4711John M1723M

arc M6381Betty Magnetic Disks
4711 John M F Access Time
1723 Marc M Sequential vs. Random

Access
6381 Betty F I/O Parallelism

Alternative Storage
• Record identifier (rid): Techniques

Solid-State Disks
hpageno, slotnoi Network-Based Storage
Managing Space
• Record position (within page): Free Space Management
slotno × bytes per slot Buffer Manager

slot no. of records
• Record deletion? directory in this page
Databases vs.
• record id should not Operating Systems
Files and Records

change Heap Files
⇒ slot directory (bitmap) 101 3 Header Free Space Management

Inside a Page
Recap
46
Storage
Inside a Page—Variable-Sized Fields
Torsten Grust
• Variable-sized fields moved to

end of each record. 4 7 1 1 • M J o h n⊥1 7 2 3 • M M
a r c⊥6 3 8 1 • F B e t t y⊥
• Placeholder points to
Magnetic Disks
location.

Access Time
• Why? Sequential vs. Random

Access
I/O Parallelism
Alternative Storage
Techniques
Solid-State Disks
Managing Space
no. of records Buffer Manager

in this page Replacement Policies
Databases vs.
Operating Systems
Files and Records

3 Header
Heap Files
Inside a Page
Recap
47
Storage
Torsten Grust

end of each record. 4 7 1 1 • M J o h n⊥1 7 2 3 • M M
a r c⊥6 3 8 1 • F B e t t y⊥
Magnetic Disks
location.

Access Time

Access
• Slot directory points to start of I/O Parallelism

each record. Alternative Storage
Techniques
Solid-State Disks
Managing Space
slot no. of records Buffer Manager

directory in this page Replacement Policies
Databases vs.
Operating Systems
• • • Files and Records

3 Header
Heap Files
Inside a Page
Recap
47
Storage
Torsten Grust

end of each record. 4 7 1 1 • M J o h n⊥1 7 2 3 MM
a r c⊥6 3 8 1 • F B e t t y⊥1 7 2
• Placeholder points to 3 • M T i m o t h y⊥ Magnetic Disks
location.

Access Time

Access

• Records can move on page. Techniques

Solid-State Disks
• E.g., if field size changes or Network-Based Storage
Managing Space
page is compacted. Free Space Management
slot no. of records Buffer Manager

directory in this page Replacement Policies
Databases vs.
Operating Systems

3 Header
Heap Files
Inside a Page
Recap
47
Storage
Torsten Grust

a r c⊥6 3 8 1 • F B e t t y⊥
• Placeholder points to •
Magnetic Disks
location.

Access Time
forward
Access


Solid-State Disks
Managing Space
• Create “forward address” if slot no. of records Buffer Manager

directory in this page
record won’t fit on page.

• Future updates? Databases vs.

Operating Systems

3 Header
Heap Files
Inside a Page
Recap
47
Storage
Torsten Grust

a r c⊥6 3 8 1 • F B e t t y⊥
Magnetic Disks
location.

Access Time

Access


Solid-State Disks
Managing Space
• Create “forward address” if slot no. of records Buffer Manager

directory in this page
record won’t fit on page.

• Future updates? Databases vs.

Operating Systems
• Related issue: space-efficient • • • 3 Header Files and Records

Heap Files
representation of NULL values. Free Space Management
Inside a Page
Recap
47
Storage
IBM DB2 Data Pages
Torsten Grust
Data page and layout details
• Support for 4 K, 8 K, 16 K, 32 K data pages in separate table

spaces. Buffer manager pages match in size. Magnetic Disks
Access Time
• 68 bytes of database manager overhead per page. On a 4 K Sequential vs. Random

Access
page: maximum of 4,028 bytes of user data (maximum I/O Parallelism
record size: 4,005 bytes). Records do not span pages. RAID Levels 1, 0, and 5
Alternative Storage
• Maximum table size: 512 GB (with 32 K pages). Maximum Techniques
Solid-State Disks

number of columns: 1,012 (4 K page: 500), maximum Network-Based Storage
number of rows/page: 255. IBM DB2 RID format? Managing Space

• Columns of type LONG VARCHAR, CLOB, etc. maintained Buffer Manager

outside regular data pages (pages contain descriptors only). Replacement Policies
• Free space management: first-fit order. Free space map Databases vs.
Operating Systems
distributed on every 500th page in FSCR (free space control Files and Records
records). Records updated in-place if possible, otherwises Heap Files
uses forward records. Inside a Page
Recap
48
Storage
IBM DB2 Data Pages
Torsten Grust
Taken directly from the DB2 V9.5 Information Center
Magnetic Disks
Access Time
Access
I/O Parallelism
Alternative Storage
Techniques
Solid-State Disks
Managing Space
Buffer Manager
Databases vs.
Operating Systems
Files and Records

Heap Files
Inside a Page
http://publib.boulder.ibm.com/infocenter/db2luw/v9r5/ Alternative Page Layouts
Recap
49
Storage
Torsten Grust
We have just populated data pages in a row-wise fashion:
a1 b1 c1 a4 b4 c4
Magnetic Disks
a1 b1 c1 d1 c1 d1 a2 c4 d4 Access Time
a2 b2 c2 d2 b2 c2 d2 Sequential vs. Random

d2 a3 b3 Access
a3 b3 c3 d3 c3 d3 I/O Parallelism
a4 b4 c4 d4 RAID Levels 1, 0, and 5
page 0 page 1
Alternative Storage
Techniques
Solid-State Disks
We could as well do that column-wise: Network-Based Storage
Managing Space
Buffer Manager
a1 a2 a3 b1 b2 b3 Replacement Policies
a1 b1 c1 d1 a3 a4 b3 b4
Databases vs.
a2 b2 c2 d2 Operating Systems
a3 b3 c3 d3 Files and Records

Heap Files
a4 b4 c4 d4 Free Space Management
page 0 page 1 Inside a Page
Recap
50
Storage
Torsten Grust
These two approaches are also known as NSM (n-ary storage

model) and DSM (decomposition storage model).1
• Tuning knob for certain workload types (e.g., OLAP)
Magnetic Disks
• Suitable for narrow projections and in-memory database Access Time
systems Access
(% Database Systems and Modern CPU Architecture) I/O Parallelism

• Different behavior with respect to compression. Alternative Storage

Techniques
Solid-State Disks
A hybrid approach is the PAX minipage 3 Managing Space
(Partition Attributes Across) layout: Free Space Management
minipage 2 Buffer Manager

• Divide each page into minipages. Pinning and Unpinning
• Group attributes into them. minipage 1 Databases vs.

Operating Systems
% Ailamaki et al. Weaving Relations for Cache minipage 0 Files and Records
Heap Files
Performance. VLDB 2001. Free Space Management
page 0
Inside a Page
Recap
1 Recently, the terms row-store and column-store have become popular, too. 51
Storage
Recap
Torsten Grust
Magnetic Disks
Random access orders of magnitude slower than Magnetic Disks
sequential. Access Time
Access
Disk Space Manager I/O Parallelism
Abstracts from hardware details and maps RAID Levels 1, 0, and 5
page number 7→ physical location. Alternative Storage

Techniques
Solid-State Disks
Buffer Manager Network-Based Storage
Page caching in main memory; pin ()/unpin () interface; Managing Space

replacement policy crucial for effectiveness.
Buffer Manager
File Organization Replacement Policies
Stable record identifiers (rids); maintenance with Databases vs.

Operating Systems
fixed-sized records and variable-sized fields; NSM vs. DSM.
Files and Records
Heap Files
Inside a Page
Recap
52

db2 01

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

db2 01

Uploaded by

Copyright:

Available Formats

Storage

Architecture and Implementation of Database Systems Access

Files and Records

Web Forms Applications SQL Interface

Lock Manager Buffer Manager

Files and Records

• Fast—but expensive and small—memory close to CPU Pinning and Unpinning

• Larger, slower memory at the periphery Databases vs.

cache. Heap Files

platter heads Solid-State Disks

disk heads on the requested track Pinning and Unpinning

• Platters (disks) steadily rotate Databases vs.

reads/writes data one block at a time Heap Files

Data blocks can only be read and written if disk heads

• This design has implications on the access time to

1 Move disk arms to desired track (seek time ts ) Solid-State Disks

Files and Records

• Seagate Cheetah 15K.7 performance characteristics:

 What is the access time to read an 8 KB data block? Alternative Storage

Files and Records

• Seagate Cheetah 15K.7 performance characteristics:

 What is the access time to read an 8 KB data block? Alternative Storage

average seek time ts = 3.40 ms Network-Based Storage

Example (Read 1 000 blocks of size 8 kB)

• sequential read of adjacent blocks:

= 3.40 ms + 2.00 ms + 50 ms + 3.2 ms ≈ 58.6 ms Alternative Storage

⇒ Sequential I/O is much faster than random I/O Replacement Policies

Files and Records

• Disk manufacturers play a number of tricks to improve

Align sector 0 of each track to avoid Access Time

rotational delay during longer Access

request scheduling Solid-State Disks

If multiple requests have to be served, choose the one Managing Space

Files and Records

Disk seek and rotational latencies have only marginally improved

But: Access Time

• Hard disk capacity grows by ≈ 50 % every year

Therefore: Network-Based Storage

Example (5 Years Ago: Seagate Barracuda 7200.7) Databases vs.

The latency penalty is hard to avoid

parallelism Sequential vs. Random

Files and Records

• Replicate data onto multiple disks:

6 7 8 9 ··· Sequential vs. Random

• Achieves I/O parallelism only for reads Replacement Policies

• This is also known as RAID 1 (mirroring without parity)

(RAID: Redundant Array of Inexpensive Disks) Free Space Management

• Distribute data over disks:

• Full I/O parallelism for read and write operations

• Also known as RAID 0 (striping without parity) Files and Records

• Distribute data and parity information over > 3 disks:

1 3 5/6 7 · · · 2 3/4 5 8 · · · 1/2 4 6 7/8 · · · Alternative Storage

• High I/O parallelism Buffer Manager

(with dual parity/RAID 6: two disks may fail) Databases vs.

data and associated parity Heap Files

• Also known as RAID 5 (striping with distributed parity) Inside a Page

Solid state disks (SSDs) have emerged as an alternative to

• SSDs provide very low-latency

• Random writes, however, are

2 Once pages have been Buffer Manager

them is almost as fast as Databases vs.

• Typical page size: 128 kB

Example (Perform block-level delete to accomodate new data pages)

What is the access time to read an 8 KB data block? Alternative Storage

What is the access time to read an 8 KB data block? Alternative Storage

What is the access time to read an 8 KB data block? Alternative Storage

What is the access time to read an 8 KB data block? Alternative Storage

Allocation of contiguous pages

Allocation of contiguous pages

Why do we need the dirty bit? Buffer Manager

Why do we need the dirty bit? Buffer Manager