Chapter 21: Parallel and Distributed Storage

Database System Concepts, 7th Ed.

Database System Concepts - 7th Edition 21.2 ©Silberschatz, Korth and Sudarshan
 Data Partitioning (1)
 In its simplest form, I/O parallelism refers to reducing the time required to
retrieve relations from disk by partitioning the relations on multiple disks, on
multiple nodes (servers).
• We focus on parallelism across nodes.
• Same techniques can be used across disks on a node.
 Two Main Approaches
• Horizontal partitioning: tuples of a relation are divided among many nodes
such that some subset of tuple resides on each node.
• Vertical partitioning: e.g. r(A,B,C,D) with primary key A into r1(A,B) and
r2(A,C,D) (discussed in Chapter 13).
• By default, the word partitioning refers to horizontal partitioning.

Database System Concepts - 7th Edition 21.3 ©Silberschatz, Korth and Sudarshan
 Data Partitioning (2)
 Partitioning techniques (number of nodes = ):
Round-robin:
Send the th tuple inserted in the relation
to node mod .
Hash partitioning:
• Choose one or more attributes as the
partitioning attributes.
• Choose hash function with range of .
• Let denote result of hash function applied
to the partitioning attribute value of a
tuple. Send tuple to node .

Database System Concepts - 7th Edition 21.4 ©Silberschatz, Korth and Sudarshan
 Data Partitioning (3)

 Range partitioning:
• Choose an attribute as the partitioning
attribute.
• A partitioning vector is chosen.
• If , then .
• Consider a tuple where is the
partitioning attribute.
 If , then goes to node .
 If , then goes to node .
 If , then goes to node .

Database System Concepts - 7th Edition 21.5 ©Silberschatz, Korth and Sudarshan
 Comparison of Partitioning Techniques (1)
 Evaluate how well partitioning techniques (round robin,
hash partitioning, range partitioning) support the following
types of data access:
1. Scanning the entire relation.
 SQL: select * from r
2. Locating a tuple associatively (point queries).
 E.g., .
 SQL: select * from r where r.A = 25
3. Locating all tuples such that the value of a given
attribute lies within a specified range (range queries).
 E.g., .
 SQL: select * from r where 10 <= r.A and r.A < 25

Database System Concepts - 7th Edition 21.6 ©Silberschatz, Korth and Sudarshan
 Comparison of Partitioning Techniques (2)
Round robin:
 Best suited for sequential scan of entire relation
on each query.
• All nodes have almost an equal number of
tuples; retrieval work is thus well balanced
between nodes.
 All queries must be processed at all nodes
Hash partitioning:
 Good for sequential access
• Assuming hash function is good, and
partitioning attributes form a key, tuples will
be equally distributed between nodes
 Good for point queries on partitioning attribute
• Can lookup single node, leaving others
available for answering other queries.
 Range queries inefficient, must be processed at
all nodes

Database System Concepts - 7th Edition 21.7 ©Silberschatz, Korth and Sudarshan
 Comparison of Partitioning Techniques (3)
Range partitioning:
 Provides data clustering by partitioning attribute
value.
• Good for sequential access
• Good for point queries on partitioning
attribute: only one node needs to be
accessed.
 For range queries on partitioning attribute, one
to a few nodes may need to be accessed
• Remaining nodes are available for other
queries.

Database System Concepts - 7th Edition 21.8 ©Silberschatz, Korth and Sudarshan
 Types of Skew
 Data-distribution skew: some nodes have many tuples,
while others may have fewer tuples. Could occur due to
• Attribute-value skew.
 Some partitioning-attribute values appear in many
tuples.
 All the tuples with the same value for the partitioning
attribute end up in the same partition.
 Can occur with range-partitioning and hash-
partitioning.
• Partition skew.
 Imbalance, even without attribute-value skew
 Badly chosen range-partition vector may assign too
many tuples to some partitions and too few to others.
 Less likely with hash-partitioning
 Execution skew can occur even without data distribution
skew
• E.g. relation range-partitioned on date, and most queries
access tuples with recent dates

Database System Concepts - 7th Edition 21.9 ©Silberschatz, Korth and Sudarshan
 Virtual Node Partitioning
 Key idea: pretend there are several times (10x to 20x) as many virtual nodes as
real nodes
• Virtual nodes are mapped to real nodes.
• Tuples are partitioned across virtual nodes.
• It can be used to support any of the partitioning techniques discussed before.
 Mapping of virtual nodes to real nodes
• Round-robin: virtual node i mapped to real node .
• Mapping table: mapping table virtual_to_real_map[] tracks which virtual node
is on which real node
 Allows skew to be handled by moving virtual nodes from more loaded
nodes to less loaded nodes.
 Both data distribution skew and execution skew can be handled.

Database System Concepts - 7th Edition 21.10 ©Silberschatz, Korth and Sudarshan