VICTORIA UNIVERSITY OF WELLINGTON

Te Whare Wananga o te Upoko o te Ika a Maui

Introduction to
Cloud Databases
Lecturer : Dr. Pavle Mogin
SWEN 432
Advanced Database Design and
Implementation

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 2

Plan for Intro to Cloud Databases

An overview of Cloud Computing
What is Cloud Computing,
The Architecture of Cloud Computing
Cloud Computing Services

Database as a Service (DaaS)

The basic features of DaaS
Distribution,
Replication
Scalability

Inter node communication

Gossip protocol
Readings: Have a look at Useful Links at the Course Home
Page

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 3

What is Cloud Computing

A major new paradigm rapidly shifting the way IT
services and tools are used
The main idea:
Provide cheep infrastructure to IT online services
How has it been achieved:
Using network connections to remote applications and services
where users need to possess only a minimum of hardware and
software
Paying for applications and services as per their usage, only

Cloud Computing is based on a subscription model

that is very similar to utility services like electricity,
gas, or water

(1)

Advanced Database Design and Implementation 2016

What is Cloud Computing

Intro to Cloud Databases 4

(2)

Users do not have to invest in any major hardware, or

any major software
They access them and use them on the cloud and
pay only for the resources they use
Users also do not need to be burdened by hardware
and software installation and maintenance
Applications and databases are stored in large server
farms or data centers owned by Google, IBM,
Microsoft, Amazon,

Advanced Database Design and Implementation 2016

Cloud Computing Services

Intro to Cloud Databases 5

(1)

Software as a Service (SaaS)

Software applications are provided to users by cloud providers where
users do not have to install them at their sites

Platform as a Service (PaaS)

A hardware or software platform (a computer, an operating system, a
programming environment, run-time libraries, ) is provided to users

Infrastructure as a Service (IaaS)

A service where users can use expensive hardware like array processor
servers, and network processors

Database as a Service
Cloud storage service where users hire storage facilities, including a
DBMS and pay only for storage space they use

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 6

Cloud Computing Services

(2)

To attain scalable performance and robust availability of

services, cloud computing vendors use:
Hardware, software, and data redundancy and
Complex techniques for managing networked platforms and data
replication

Scalability means that performance of a service does

not depend on the number of users and clients
requesting the service
Availability of a service relates to the fact that the service
is always ready for use
A service is highly available if it has a small latency

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 7

Cloud Databases
Cloud databases have been developed to serve the
massive growth of digital data and consumer services
in the last 10 to 15 years:
Social networks (Facebook, Twitter),
Data storage and sync services (Dropbox, iCloud),
Desktop replacement (Google Documents)

Cloud databases are distributed database systems

that are accessed via cloud services
Database data are replicated and stored on multiple independent
servers (nodes) to achieve scalability and availability
End users and client application use data via APIs that hide details
of the exact location where an data object might be stored

Advanced Database Design and Implementation 2016

Common Features of CDBs

Intro to Cloud Databases 8

(1)

Most database services offer web-based consoles,

which the end user can use to provision and configure
database instances
Database services consist of a database manager
(DBMS) component, which controls the underlying
database instances using a service API
The service API is exposed to the end users, and permits users to
perform maintenance and scaling operations on their database
instances

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 9

Common Features of CDBs

(2)

Database services make the underlying software

stack transparent to the user
The stack includes an OS, DBMS and third-party software used by
the database,
The service provider is responsible for installing, patching and
updating the underlying software stack

Database services take care of scalability and high

availability of the database
Some vendors offer auto-scaling, others enable users to scale up
using an API
There is typically a commitment for a certain level of high availability
(e.g. 99.9% or 99.99% of the time).

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 10

Network Architectures of Cloud DBs

Large cloud databases distributed over several node

machines fall under one of the following two
categories:
Shared nothing, where each node contains a database partition
and whole responsibility for data it holds,
Shared disk, where all nodes have access to shared disks
containing all database data and all nodes share responsibility for
the sole copy of the database

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 11

Shared Nothing and Shared Disk

In a shared nothing architecture:
A middleware is required to route incoming data requests to the
node(s) responsible for data requested,
Reads and Writes involving data on a single node are efficient,
Reads and Writes involving data on multiple nodes are inefficient,
since joins and constraints must be validated over multiple nodes,
Nodes can be added, or removed easily without affecting other nodes,
making the architecture scalable
Very suitable for using commodity machines

In a shared disk architecture:

All nodes have access to the entire database (no middleware
required),
Allow for high database consistency needed by OLTP
Locking (for consistency) and logging (for recovery) introduce
overheads, so hard to scale

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 12

Data Replication and Distribution

Practically all CDBMSs replicate data on several
machines
Several copies of the same data are available to many users
accessing the database simultaneously
Data availability has been enhanced in the case of: a great number
of users accessing same data, or in the case of a server or network
failure

Machines, used to store replicas of a database may

belong to several data centers
A data center contains:
Servers, telecommunication infrastructure, buck-up and security
facilities
Users are guided to their databases by cloud DBMS APIs and
remain unaware of the exact location of their data

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 13

Machine Layout
The underlying infrastructure is usually composed of a
large number of networked commodity machines
Each machine is called a physical node (PN)
Each PN has the same software configuration but
may have varying hardware performance
Processor speed,
Memory and disk capacity

Depending on its performance, each PN contains a

number of virtual nodes

Advanced Database Design and Implementation 2016

Physical and Virtual Nodes

Physical Node
Virtual Node
Virtual Node
Virtual Node

Network

Intro to Cloud Databases 14

Advanced Database Design and Implementation 2016

Scalability

Intro to Cloud Databases 15

(1)

Scalability is the ability of a system:

To handle a growing amount of work in a capable manner or
The ability to be enlarged to accommodate the workload growth
This can refer to the capability of a system to increase total
throughput under an increased load when resources (typically
hardware) are added
A system whose performance improves after adding hardware
proportionally to the capacity added, is said to be a scalable
system
If the system fails when the workload or the quantity of hardware
added increase, it does not scale

One of the primary goals of cloud database systems

is achieving cost effective scalability

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 16

Scalability

(2)

In principle, scaling of cloud databases can be

achieved by:
Dedicating more nodes to the database system (referred to as
horizontal scaling), and
Adding more resources (memory, CPU) to individual existing nodes

Horizontal scaling is important to cloud databases

because of its cost effectiveness
A horizontally scalable cloud database can be run on cheaper
commodity hardware
As the number of users and data grow and more performance is
required, more cheap nodes are added, and data and work load are
distributed to the new nodes

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 17

Node Communication in a Network

In a networked and replicated database system, where
a client may issue a read or update request to any
node, nodes have to communicate in order to:
Dispatch the clients request to a corresponding replica and
Propagate the clients updates to all replicas

Very often, nodes communicate using a gossip

protocol
Gossip protocols are inspired by the form of gossip seen in social
networks

The term epidemic protocol is sometimes used as a

synonym for a gossip protocol, because gossip
spreads information in a manner similar to the spread
of a virus in biological communities

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 18

Gossip Protocol
The gossip protocol is an inter node communication
protocol that satisfies the following conditions:
1. The core of the protocol involves periodic, pair wise inter node
interactions
2. The information exchange during these interactions is of a limited
size
3. When nodes interact, the state of at least one of them changes to
reflect the state of the other one
4. Reliable communication is not assumed
5. The frequency of the interaction is low compared to typical
message latency, so that the cost of the protocol is negligible
6. There is some form of randomness in the peer selection
Peers may be selected from the full set of nodes or from a
smaller set of nodes (nodes hosting a replica of the same data)

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 19

Efficiency of a Gossip Protocol

This is an approximate calculation
Assume, in the first round of gossiping, a node gets an information
that it needs to share with others
So, the node picks another node and after the second round of
gossiping two nodes know for the information
In the third round each node knowing the information shares it with two
new nodes and that results in four nodes knowing the information
After the round i there are 2 (i 1) nodes knowing the information
Assume after the round h all n nodes in the system know the
information

n 2 h-1
The number of rounds to disseminate information to all n nodes is

h (log

2n )1

Advanced Database Design and Implementation 2016

Intro to Cloud Databases 20

Gossip Protocol (Example)

Assume:
A network contains n = 25 000 nodes
A gossip occurs every 100 ms

Then:
There are h = 15 rounds needed to spread information through the
network and it takes 1.5 seconds

Advanced Database Design and Implementation 2016

Summary

Intro to Cloud Databases 21

(1)

Cloud Computing:
Storing and accessing applications and computer data through a Web
browser rather than running installed software on your computer
Internet-based computing whereby information, IT resources, and
software applications are provided to computers and mobile devices
on demand

Cloud Computing Services leverage:

Commodity hardware,
Data redundancy, and
Robust availability

over a collection of networked computers and reduce

complexity of managing such systems by abstracting
implementation details to a user

Advanced Database Design and Implementation 2016

Summary

Intro to Cloud Databases 22

(2)

Cloud databases have been developed to serve the

massive growth of digital data and consumer services
in last 10 to 15 years
The database service provider takes responsibility for
installing and maintaining the database, and
application owners pay according to their usage.
Many cloud databases use the shared nothing
architecture and replicate data on several network
nodes
Scalability is the ability of a system to increase total
throughput under an increased load when resources
(typically hardware) are added

Advanced Database Design and Implementation 2016

Summary

Intro to Cloud Databases 23

(3)

Gossip inter node communication protocols are very

popular among cloud database management systems
They spread information and exchange states in a
very reliable way, since there is no single point of
failure
There are approximately log2 n rounds of gossiping
needed to spread an information through a network of
n nodes