Classification of Physical Storage Media

Speed with which data can be accessed

Cost per unit of data
Reliability
data loss on power failure or system crash
physical failure of the storage device
Can differentiate storage into:
o volatile storage: loses contents when power is switched off
o nonvolatile storage: Contents persist even when power is switched off.
Includes secondary and tertiary storage, as well as
battery backed up main memory.

Physical Storage Media

Cache fastest and most costly form of storage; volatile; managed by the computer
system hardware (Note: Cache is pronounced as cash)
Main memory:
fast access (10s to 100s of nanoseconds; 1 nanosecond = 109 seconds)
generally too small (or too expensive) to store the entire database
capacities of up to a few Gigabytes widely used currently
Capacities have gone up and perbyte costs have decreased steadily and
rapidly (roughly factor of 2 every 2 to 3 years)
Volatile contents of main memory are usually lost if a power failure or system crash
occurs.
Flash memory
Data survives power failure
Data can be written at a location only once, but location can be erased and
written to again
Can support only a limited number (10K 1M) of write/erase cycles.
Erasing of memory has to be done to an entire bank of memory
Reads are roughly as fast as main memory
But writes are slow (few microseconds), erase is slower
Flash memory
NOR Flash
Fast reads, very slow erase, lower capacity
Used to store program code in many embedded devices
NAND Flash
Page at a time read/write, multipage erase
High capacity (several GB)
Widely used as data storage mechanism in portable devices

Magnetic disk
Data is stored on spinning disk, and read/written magnetically
Primary medium for the long term storage of data; typically stores entire database.
Data must be moved from disk to main memory for access, and written back for storage
directaccess
possible to read data on disk in any order, unlike magnetic tape

Survives power failures and system crashes

disk failure can destroy data: is rare but does happen
Optical storage
Nonvolatile, data is read optically from a spinning disk using a laser
CDROM
(640 MB) and DVD (4.7 to 17 GB) most popular forms
Write one, read many
(WORM) optical disks used for archival storage (CDR, DVDR, DVD+R)
Multiple write versions also available (CDRW, DVDRW, DVD+RW, and DVDRAM)
Reads and writes are slower than with magnetic disk
Jukebox
systems, with large numbers of removable disks, a few drives, and a mechanism for
automatic loading/unloading of disks available for storing large volumes of data
Tape storage
nonvolatile, used primarily for backup (to recover from disk failure), and for archival
data
sequentialaccess
much slower than disk
very high capacity (40 to 300 GB tapes available)
tape can be removed from drive storage costs much cheaper than disk, but drives are
expensive
Tape jukeboxes available for storing massive amounts of data
hundreds of terabytes (1 terabyte = 109 bytes) to even a petabyte (1 petabyte =
1012 bytes)

Storage Hierarchy
primary storage: Fastest media but volatile (cache, main memory).
secondary storage: next level in hierarchy, nonvolatile, moderately fast access time
also called online storage
E.g. flash memory, magnetic disks
tertiary storage: lowest level in hierarchy, nonvolatile, slow access time
also called offline storage
E.g. magnetic tape, optical storage

Storage Hierarchy

Magnetic Hard Disk Mechanism

NOTE: Diagram is schematic, and simplifies the structure of actual disk drives

Magnetic Disks
Readwrite head
Positioned very close to the platter surface (almost touching it)
Reads or writes magnetically encoded information.
Surface of platter divided into circular tracks
Over 50K100K tracks per platter on typical hard disks
Each track is divided into sectors.
Sector size typically 512 bytes
Typical sectors per track: 500 (on inner tracks) to 1000 (on outer tracks)
To read/write a sector
disk arm swings to position head on right track
platter spins continually; data is read/written as sector passes under head
Headdisk assemblies
multiple disk platters on a single spindle (1 to 5 usually)
one head per platter, mounted on a common arm.
Cylinder i consists of i track of all the platters
Earlier generation disks were susceptible to head crashes leading to loss of all
data on disk
Current generation disks are less susceptible to such disastrous failures, but
individual sectors may get corrupted
th

Disk Controller
Disk controller interfaces between the computer system and the disk drive hardware.
accepts highlevel commands to read or write a sector
initiates actions such as moving the disk arm to the right track and actually
reading or writing the data
Computes and attaches checksums to each sector to verify that data is read back
correctly
If data is corrupted, with very high probability stored checksum wont
match recomputed checksum
Ensures successful writing by reading back sector after writing it
Performs remapping of bad sectors

Disk Subsystem

Disk interface standards families

ATA (AT adaptor) range of standards
SATA (Serial ATA)
SCSI (Small Computer System Interconnect) range of standards
Several variants of each standard (different speeds and capabilities)

Performance Measures of Disks

Access time the time it takes from when a read or write request is issued to when data
transfer begins. Consists of:
Seek time time it takes to reposition the arm over the correct track.
Average seek time is 1/2 the worst case seek time.
Would be 1/3 if all tracks had the same number of sectors, and we
ignore the time to start and stop arm movement
4 to 10 milliseconds on typical disks
Rotational latency time it takes for the sector to be accessed to appear under the
head.
Average latency is 1/2 of the worst case latency.
4 to 11 milliseconds on typical disks (5400 to 15000 r.p.m.)
Data transfer rate the rate at which data can be retrieved from or stored to the disk.
25 to 100 MB per second max rate, lower for inner tracks
Multiple disks may share a controller, so rate that controller can handle is also
important
E.g. ATA5: 66 MB/sec, SATA: 150 MB/sec, Ultra 320 SCSI: 320 MB/s
Fiber Channel (FC2Gb): 256 MB/s
Mean time to failure (MTTF) the average time the disk is expected to run continuously
without any failure.
Typically 3 to 5 years
Probability of failure of new disks is quite low, corresponding to a theoretical MTTF of
500,000 to 1,200,000 hours for a new disk
E.g., an MTTF of 1,200,000 hours for a new disk means that given 1000
relatively new disks, on an average one will fail every 1200 hours
MTTF decreases as disk ages

Optimization of Disk Block Access

Block a contiguous sequence of sectors from a single track
data is transferred between disk and main memory in blocks
Typical block sizes today range from 4 to 16 kilobytes
Diskarmscheduling algorithms order pending accesses to tracks so that disk arm
movement is minimized
elevator algorithm : move disk arm in one direction (from outer to inner tracks or vice
versa), processing next request in that direction, till no more requests in that direction, then
reverse direction and repeat
File organization optimize block access time by organizing the blocks to correspond
to how data will be accessed
E.g. Store related information on the same or nearby blocks/cylinders.

File systems attempt to allocate contiguous chunks of blocks (e.g. 8 or 16

blocks) to a file
Files may get fragmented over time
E.g. if data is inserted to/deleted from the file
Or free blocks on disk are scattered, and newly created file has its
blocks scattered over the disk
Sequential access to a fragmented file results in increased disk arm
movement
Some systems have utilities to defragment the file system, in order to speed up
file access
Nonvolatile write buffers speed up disk writes by writing blocks to a nonvolatile
RAM buffer immediately
Nonvolatile
RAM: battery backed up RAM or flash memory
Even if power fails, the data is safe and will be written to disk when
power returns
Controller then writes to disk whenever the disk has no other requests or request has
been pending for some time
Database operations that require data to be safely stored before continuing can continue
without waiting for data to be written to disk
Writes can be reordered to minimize disk arm movement
Log disk a disk devoted to writing a sequential log of block updates
Used exactly like nonvolatile RAM
Write to log disk is very fast since no seeks are required
No need for special hardware (NVRAM)
File systems typically reorder writes to disk to improve performance
Journaling file systems write data in safe order to NVRAM or log disk
Reordering without journaling: risk of corruption of file system data

RAID
RAID: Redundant Arrays of Independent Disks
disk organization techniques that manage a large numbers of disks, providing a view of
a single disk of
high capacity and high speed by using multiple disks in parallel, and
high reliability by storing data redundantly, so that data can be recovered even
if a disk fails
The chance that some disk out of a set of N disks will fail is much higher than the
chance that a specific single disk will fail.
E.g., a system with 100 disks, each with MTTF of 100,000 hours (approx. 11 years),
will have a system MTTF of 1000 hours (approx. 41 days)

Improvement of Reliability via Redundancy

Redundancy store extra information that can be used to rebuild information lost in a
disk failure
E.g., Mirroring (or shadowing)
Duplicate every disk. Logical disk consists of two physical disks.
Every write is carried out on both disks

Reads can take place from either disk

If one disk in a pair fails, data still available in the other
Data loss would occur only if a disk fails, and its mirror disk also fails
before the system is repaired
Probability of combined event is very small
Except for dependent failure modes such as fire or building collapse or
electrical power surges

Improvement of Reliability via Redundancy

Mean time to data loss depends on mean time to failure, and mean time to repair
E.g. MTTF of 100,000 hours, mean time to repair of 10 hours gives mean time
to data loss of 500*106 hours (or 57,000 years) for a mirrored pair of disks (ignoring
dependent failure modes)

Improvement in Performance via Parallelism

Two main goals of parallelism in a disk system:
1. Load balance multiple small accesses to increase throughput
2. Parallelize large accesses to reduce response time.
Improve transfer rate by striping data across multiple disks.
Bitlevel striping split the bits of each byte across multiple disks
But seek/access time worse than for a single disk
Bit level striping is not used much any more
Blocklevel striping with n disks, block i of a file goes to disk (I mod n) + 1
Requests for different blocks can run in parallel if the blocks reside on different
disks
A request for a long sequence of blocks can utilize all disks in parallel

RAID Levels
RAID organizations, or RAID levels, have differing cost, performance and reliability
characteristics
RAID Level 0: Block striping; nonredundant.
Used in high performance applications where data lost is not critical.
RAID Level 1: Mirrored disks with block striping
Offers best write performance.
Popular for applications such as storing log files in a database system.
RAID Level 2: MemoryStyle ErrorCorrectingCodes (ECC) with bit striping.
RAID Level 3: BitInterleaved Parity
a single parity bit is enough for error correction, not just detection
When writing data, corresponding parity bits must also be computed
and written to a parity bit disk
To recover data in a damaged disk, compute XOR of bits from other
disks (including parity bit disk) n RAID Level 3 (Cont.)
Faster data transfer than with a single disk, but fewer I/Os per second since
every disk has to participate in every I/O.
RAID Level 4: Block Interleaved Parity; uses blocklevel striping, and keeps a parity
block on a separate disk for corresponding blocks from N other disks.
When writing data block, corresponding block of parity bits must also be
computed and written to parity disk

To find value of a damaged block, compute XOR of bits from corresponding

blocks (including parity block) from other disks.
Provides higher I/O rates for independent block reads than Level 3
block read goes to a single disk, so blocks stored on different disks can
be read in parallel
Before writing a block, parity data must be computed
Can be done by using old parity block, old value of current block and
new value of current block (2 block reads + 2 block writes)

Or by recomputing the parity value using the new values of blocks

corresponding to the parity block
More efficient for writing large amounts of data sequentially
Parity block becomes a bottleneck for independent block writes since every
block write also writes to parity disk
RAID Level 5: Block Interleaved Distributed Parity; partitions data and parity among
all N + 1 disks, rather than storing data in N disks and parity in 1 disk.
E.g., with 5 disks, parity block for nth set of blocks is stored on disk (n mod 5) + 1,
with the data blocks stored on the other 4 disks.

Higher I/O rates than Level 4.

Block writes occur in parallel if the blocks and their parity blocks are on
different disks.
Subsumes Level 4: provides same benefits, but avoids bottleneck of parity disk.
RAID Level 6: P+Q Redundancy scheme; similar to Level 5, but stores extra redundant
information to guard against multiple disk failures.
Better reliability than Level 5 at a higher cost; not used as widely.

Choice of RAID Level

Factors in choosing RAID level
Monetary cost
Performance: Number of I/O operations per second, and bandwidth during
normal operation
Performance during failure
Performance during rebuild of failed disk
Including time taken to rebuild failed disk
RAID 0 is used only when data safety is not important
E.g. data can be recovered quickly from other sources
Level 2 and 4 never used since they are subsumed by 3 and 5
Level 3 is not used since bit striping forces single block reads to access all disks,
wasting disk arm movement
Level 6 is rarely used since levels 1 and 5 offer adequate safety for most applications
So competition is mainly between 1 and 5
Level 1 provides much better write performance than level 5
Level 5 requires at least 2 block reads and 2 block writes to write a single block,
whereas Level 1 only requires 2 block writes
Level 1 preferred for high update environments such as log disks
Level 1 had higher storage cost than level 5

disk drive capacities increasing rapidly (50%/year) whereas disk access times
have decreased much less (x 3 in 10 years)
I/O requirements have increased greatly, e.g. for Web servers
When enough disks have been bought to satisfy required rate of I/ O, they often
have spare storage capacity so there is often no extra monetary cost for Level 1!
Level 5 is preferred for applications with low update rate, and large amounts of data
Level 1 is preferred for all other applications

Hardware Issues
Software RAID: RAID implementations done entirely in software, with no special
hardware support
Hardware RAID: RAID implementations with special hardware
Use nonvolatile RAM to record writes that are being executed
Beware: power failure during write can result in corrupted disk
E.g. failure after writing one block but before writing the second in a
mirrored system
Such corrupted data must be detected when power is restored
Recovery from corruption is similar to recovery from failed
disk
NVRAM helps to efficiently detected potentially corrupted
blocks
Otherwise all blocks of disk must be read and compared with
mirror/parity block
Hot swapping: replacement of disk while system is running, without power down
Supported by some hardware RAID systems,
reduces time to recovery, and improves availability greatly
Many systems maintain spare disks which are kept online, and used as replacements for
failed disks immediately on detection of failure
Reduces time to recovery greatly
Many hardware RAID systems ensure that a single point of failure will not stop the
functioning of the system by using
Redundant power supplies with battery backup
Multiple controllers and multiple interconnections to guard against
controller/interconnection failures

RAID Terminology in the Industry

RAID terminology not very standard in the industry
E.g. Many vendors use
RAID 1: for mirroring without striping
RAID 10 or RAID 1+0: for mirroring with striping
Hardware RAID implementations often just offload RAID processing onto a
separate subsystem, but dont offer NVRAM.
Read the specs carefully!
Software RAID supported directly in most operating systems today

Optical Disks
Compact diskread only memory (CDROM)
Seek time about 100 msec (optical read head is heavier and slower)
Higher latency (3000 RPM) and lower data transfer rates (36 MB/s) compared
to magnetic disks
Digital Video Disk (DVD)
DVD5 holds 4.7 GB , variants up to 17 GB
Slow seek time, for same reasons as CDROM
Record once versions (CDR and DVDR)

Magnetic Tapes
Hold large volumes of data and provide high transfer rates
Few GB for DAT (Digital Audio Tape) format, 1040 GB with DLT (Digital
Linear Tape) format, 100 400 GB+ with Ultrium format, and 330 GB with Ampex
helical scan format
Transfer rates from few to 10s of MB/s
Currently the cheapest storage medium
Tapes are cheap, but cost of drives is very high
Very slow access time in comparison to magnetic disks and optical disks
limited to sequential access.
Some formats (Accelis) provide faster seek (10s of seconds) at cost of lower
capacity
Used mainly for backup, for storage of infrequently used information, and as an offline
medium for transferring information from one system to another.
Tape jukeboxes used for very large capacity storage
(terabyte (10 bytes) to petabye (10 bytes)
12

15

Storage Access
A database file is partitioned into fixed length storage units called blocks. Blocks are
units of both storage allocation and data transfer.
Database system seeks to minimize the number of block transfers between the disk and
memory. We can reduce the number of disk accesses by keeping as many blocks as possible in
main memory.
Buffer portion of main memory available to store copies of disk blocks.
Buffer manager subsystem responsible for allocating buffer space in main memory.

File Organization
The database is stored as a collection of files. Each file is a sequence of records. A
record is a sequence of fields.
One approach:
assume record size is fixed
each file has records of one particular type only
different files are used for different relations
This case is easiest to implement; will consider variable length records later.

Fixed Length Records

Simple approach:
Store record i starting from byte n * (i 1), where n is the size of each record.
Record access is simple but records may cross blocks
Modification: do not allow records to cross block boundaries
Deletion of record i: alternatives:
move records i + 1, . . ., n to i, . . . , n 1
move record n to i
do not move records, but link all free records on a free list

Free Lists
Store the address of the first deleted record in the file header.
Use this first record to store the address of the second deleted record, and so on
Can think of these stored addresses as pointers since they point to the location of a
record.
More space efficient representation: reuse space for normal attributes of free records to
store pointers. (No pointers stored in inuse records.)

Variable Length Records

Variable length records arise in database systems in several ways:
Storage of multiple record types in a file.
Record types that allow variable lengths for one or more fields.
Record types that allow repeating fields (used in some older data models)

Variable Length Records: Slotted Page Structure

Slotted page header contains:

number of record entries
end of free space in the block
location and size of each record
Records can be moved around within a page to keep them contiguous with no empty
space between them; entry in the header must be updated.
Pointers should not point directly to record instead they should point to the entry for
the record in header.

Organization of Records in Files

Heap a record can be placed anywhere in the file where there is space
Sequential store records in sequential order, based on the value of the search key of
each record
Hashing a hash function computed on some attribute of each record; the result
specifies in which block of the file the record should be placed
Records of each relation may be stored in a separate file. In a multitable clustering file
organization records of several different relations can be stored in the same file
Motivation: store related records on the same block to minimize I/O

Sequential File Organization

Suitable for applications that require sequential processing of the entire file
The records in the file are ordered by a searchkey
Deletion use pointer chains
Insertion locate the position where the record is to be inserted
if there is free space insert there l if no free space, insert the record in an
overflow block
In either case, pointer chain must be updated
Need to reorganize the file from time to time to restore sequential order

Multitable Clustering File Organization

Store several relations in one file using a multitable clustering file organization
Multitable clustering organization of customer and depositor:
good for queries involving depositor customer, and for queries involving one
single customer and his accounts
bad for queries involving only customer

Indexing and Hashing

Basic Concepts
Indexing mechanisms used to speed up access to desired data.
E.g., author catalog in library
Search Key attribute to set of attributes used to look up records in a file.
An index file consists of records (called index entries) of the form
Index files are typically much smaller than the original file
Two basic kinds of indices:
Ordered indices: search keys are stored in sorted order
Hash indices: search keys are distributed uniformly across buckets using a
hash function.

Index Evaluation Metrics

Access types supported efficiently. E.g.,
records with a specified value in the attribute
or records with an attribute value falling in a specified range of values (e.g.
10000 < salary < 40000)
Access time
Insertion time
Deletion time
Space overhead

Ordered Indices
In an ordered index, index entries are stored sorted on the search key value. E.g.,
author catalog in library.
Primary index: in a sequentially ordered file, the index whose search key specifies the
sequential order of the file.
Also called clustering index
The search key of a primary index is usually but not necessarily the primary
key.
Secondary index: an index whose search key specifies an order different from the
sequential order of the file. Also called nonclustering index.
Indexsequential file: ordered sequential file with a primary index.

Dense Index Files

Dense index Index record appears for every searchkey value in the file.

Sparse Index Files

Sparse Index: contains index records for only some searchkey values.
Applicable when records are sequentially ordered on searchkey
To locate a record with searchkey value K we:
Find index record with largest searchkey value < K
Search file sequentially starting at the record to which the index record points
Compared to dense indices:
Less space and less maintenance overhead for insertions and deletions.
Generally slower than dense index for locating records.
Good tradeoff: sparse index with an index entry for every block in file, corresponding
to least searchkey value in the block.

Multilevel Index
If primary index does not fit in memory, access becomes expensive.
Solution: treat primary index kept on disk as a sequential file and construct a sparse
index on it.
outer index a sparse index of primary index
inner index the primary index file
If even outer index is too large to fit in main memory, yet another level of index can be
created, and so on.

Indices at all levels must be updated on insertion or deletion from the file.

Index Update: Record Deletion

If deleted record was the only record in the file with its particular searchkey value, the
searchkey is deleted from the index also.

Singlelevel index deletion:

Dense indices deletion of searchkey: similar to file record deletion.
Sparse indices
if deleted key value exists in the index, the value is replaced by the next
searchkey value in the file (in searchkey order).
If the next searchkey value already has an index entry, the entry is
deleted instead of being replaced.

Index Update: Record Insertion

Singlelevel index insertion:
Perform a lookup using the key value from inserted record
Dense indices if the searchkey value does not appear in the index, insert it.
Sparse indices if index stores an entry for each block of the file, no change
needs to be made to the index unless a new block is created.
If a new block is created, the first searchkey value appearing in the
new block is inserted into the index.
Multilevel insertion (as well as deletion) algorithms are simple extensions of the
singlelevel algorithms

Secondary Indices Example

Secondary index on balance field of account

Index record points to a bucket that contains pointers to all the actual records with that
particular searchkey value.
Secondary indices have to be dense

Primary and Secondary Indices

Indices offer substantial benefits when searching for records.
BUT: Updating indices imposes overhead on database modification when
a file is modified, every index on the file must be updated,
Sequential scan using primary index is efficient, but a sequential scan using a secondary
index is expensive
Each record access may fetch a new block from disk
Block fetch requires about 5 to 10 micro seconds, versus about 100 nanoseconds
for memory access

B Tree Index Files

+

n Disadvantage

of indexedsequential
files
l performance degrades as file grows, since many overflow
blocks get created.
l Periodic reorganization of entire file is required.
n Advantage of B tree
index files:
l automatically reorganizes itself with small, local, changes,
in the face of insertions and deletions.
l Reorganization of entire file is not required to maintain
performance.
n (Minor) disadvantage of B trees:
l extra insertion and deletion overhead, space overhead.
n Advantages of B trees
outweigh disadvantages
l B trees
are used extensively
+

A B tree is a rooted tree satisfying the following properties:

+

n All paths from root to leaf are of the same length

n Each node that is not a root or a leaf has between n/2and n
children.
n A leaf node has between (n1)/2and n1 values
n Special cases:
l If the root is not a leaf, it has at least 2 children.
l If the root is a leaf (that is, there are no other nodes in the
tree), it can have between 0 and (n1) values.

B Tree Node Structure

+

Typical node

K are the searchkey

values
l P are pointers to children (for nonleaf
nodes) or pointers to
i

records or buckets of records (for leaf nodes).

n The searchkeys
in a node are ordered
K<K<K<...<K
1

n1

Leaf Nodes in B+Trees

Properties of a leaf node:
For i = 1, 2, . . ., n1, pointer P either points to a file record with
searchkey
value K , or to a bucket of pointers to file records,
each record having searchkey
value K . Only need bucket
structure if searchkey
does not form a primary key.
n If L , L are leaf nodes and i < j, L s searchkey
values are less
than L s searchkey
values
n P points to next leaf node in searchkey
order
i

NonLeaf
Nodes in B+Trees
Non leaf nodes form a multilevel
sparse index on the leaf
nodes. For a nonleaf
node with m pointers:
l All the searchkeys
in the subtree to which P points are
less than K
l For 2 i n
1, all the searchkeys
in the subtree to
which P points have values greater than or equal to K
1

i1

and less than K

l All the searchkeys
in the subtree to which P points have
values greater than or equal to K
i

n1

Example of a B+tree

B tree
for account file (n = 3)
+

Example of B+tree
n Leaf

nodes must have between 2 and 4 values

((n1)/2and n 1, with n = 5).
n Nonleaf
nodes other than root must have between 3
and 5 children ((n/2and n with n =5).
n Root must have at least 2 children.

B tree
for account file (n = 5)
+

Observations about B+trees

n Since

the internode
connections are done by pointers,
logically close blocks need not be physically close.
n The nonleaf
levels of the B tree
form a hierarchy of sparse
indices.
n The B tree
contains a relatively small number of levels
+

Level below root has at least 2* n/2 values

Next level has at least 2* n/2 * n/2 values
.. etc.
l If there are K searchkey
values in the file, the tree height is
no more than log (K)
l thus searches can be conducted efficiently.
n Insertions and deletions to the main file can be handled
efficiently, as the index can be restructured in logarithmic time.

n/2

Queries on B+Trees
Find all records with a searchkey
value of k.
H N=root
H Repeat
4 Examine N for the smallest searchkey
value > k.
4 If such a value exists, assume it is K . Then set N = P
4 Otherwise k K . Set N = P
Until N is a leaf node
H If for some i, key K = k follow pointer P to the desired record or bucket.
H Else no record with searchkey
value k exists.
i

n1

If there are K searchkey

values in the file, the height of the
tree is no more than log (K) .
n A node is generally the same size as a disk block, typically 4
kilobytes
l and n is typically around 100 (40 bytes per index entry).
n With 1 million search key values and n = 100
l at most log (1,000,000) = 4 nodes are accessed in a
lookup.
n Contrast this with a balanced binary tree with 1 million search
key values around 20 nodes are accessed in a lookup
l above difference is significant since every node access
may need a disk I/O, costing around 20 milliseconds
n/2

50

Updates on B+Trees:

Insertion
1. Find

the leaf node in which the searchkey

value would appear
2. If the searchkey
value is already present in the leaf node
n Add record to the file
3. If the searchkey
value is not present, then
1. add the record to the main file (and create a bucket if
necessary)
2. If there is room in the leaf node, insert (keyvalue,
pointer)
pair in the leaf node
3. Otherwise, split the node (along with the new (keyvalue,
pointer) entry) as discussed in the next slide. Splitting a leaf node:
l take

the n (searchkey
value, pointer) pairs (including the one
being inserted) in sorted order. Place the first n/2 in the original
node, and the rest in a new node.
l let the new node be p, and let k be the least key value in p. Insert
(k,p) in the parent of the node being split.
l If the parent is full, split it and propagate the split further up.
n Splitting of nodes proceeds upwards till a node that is not full is found.
l In the worst case the root node may be split increasing the height
of the tree by 1.

Result of splitting node containing Brighton and Downtown on inserting

Clearview
Next step: insert entry with (Downtown,pointertonewnode)
into parent

B Tree
before and after insertion of Clearview
+

Splitting a nonleaf
node: when inserting (k,p) into an already
full internal node N
l Copy N to an inmemory
area M with space for n+1
pointers and n keys
l Insert (k,p) into M
l Copy P ,K , , K
,P from M back into node N
l Copy P
,K ,,K ,P from M into newly
allocated node N
l Insert (K
,N) into parent N
1

n/2 1

n/2

n/2 +1

n/2 +1

n+1

n/2

Updates on B+Trees:
Deletion
n Find

the record to be deleted, and remove it from the main file

and from the bucket (if present)
n Remove (searchkey
value, pointer) from the leaf node if there
is no bucket or if the bucket has become empty
n If the node has too few entries due to the removal, and the
entries in the node and a sibling fit into a single node, then
merge siblings:
l Insert all the searchkey
values in the two nodes into a
single node (the one on the left), and delete the other node.
l Delete the pair (K , P ), where P is the pointer to the
deleted node, from its parent, recursively using the above
procedure.
Otherwise, if the node has too few entries due to the removal,
but the entries in the node and a sibling do not fit into a single
node, then redistribute pointers:
l Redistribute the pointers between the node and a sibling
such that both have more than the minimum number of
i1

entries.
l Update the corresponding searchkey
value in the parent of
the node.
n The node deletions may cascade upwards till a node which has
n/2 or more pointers is found.
n If the root node has only one pointer after deletion, it is deleted
and the sole child becomes the root.

Examples of B+Tree
Deletion

Before and after deleting Downtown

n Deleting Downtown causes merging of underfull
leaves
l leaf node can become empty only for n=3!

Before and After deletion of Perryridge from result of

previous example

Leaf with Perryridge becomes underfull (actually empty, in this

special case) and merged with its sibling.
n As a result Perryridge nodes parent became underfull, and was
merged with its sibling
l Value separating two nodes (at parent) moves into merged node
l Entry deleted from parent
n Root node then has only one child, and is deleted

Before and after deletion of Perryridge from earlier example

Parent of leaf containing Perryridge became underfull, and borrowed a
pointer from its left sibling
n Searchkey
value in the parents parent changes as a result

B+Tree
File Organization
n Index

file degradation problem is solved by using B Tree

indices.
n Data file degradation problem is solved by using B Tree
File
Organization.
n The leaf nodes in a B tree
file organization store records, instead
of pointers.
n Leaf nodes are still required to be half full
l Since records are larger than pointers, the maximum number
of records that can be stored in a leaf node is less than the
number of pointers in a nonleaf node.
n Insertion and deletion are handled in the same way as insertion
and deletion of entries in a B tree
index.
+

Example of B tree
File Organization
+

Good space utilization important since records use more space than
pointers.
n To improve space utilization, involve more sibling nodes in
redistribution during splits and merges
l Involving 2 siblings in redistribution (to avoid split / merge where
possible) results in each node having at least 2n / 3entries

Indexing Strings
n Variable

length strings as keys

fanout
l Use space utilization as criterion for splitting, not number of
pointers
n Prefix compression
l Key values at internal nodes can be prefixes of full key
Keep enough characters to distinguish entries in the
subtrees separated by the key value
E.g. Silas and Silberschatz can be separated by
Silb
l Keys in leaf node can be compressed by sharing common
prefixes
l Variable

BTree
Index Files
n Similar

to B+tree,

but Btree
allows searchkey
values
to appear only once; eliminates redundant storage of
search keys.
n Search keys in nonleaf nodes appear nowhere else in
the Btree;
an additional pointer field for each search
key in a nonleaf node must be included.
n Generalized Btree
leaf node

Nonleaf node pointers Bi are the bucket or file record

pointers.

Btree
(above) and B+tree
(below) on same data

BTree
Index Files (Cont.)
n Advantages

of BTree
indices:
l May use less tree nodes than a corresponding B Tree.
l Sometimes possible to find searchkey
value before reaching
leaf node.
n Disadvantages of BTree
indices:
l Only small fraction of all searchkey
values are found early
l Nonleaf
nodes are larger, so fanout
is reduced. Thus, BTrees
typically have greater depth than corresponding B Tree
l Insertion and deletion more complicated than in B Trees
+

l Implementation

is harder than B Trees.

n Typically, advantages of BTrees
do not out weigh disadvantages.
+