6 Advantages of Cloud Computing, Availability and Programming Models: Exam Guide

"Six advantages of cloud computing", "availability in cloud computing" and "cloud programming models" are three questions that belong together because they come from the same few units. In TU BSc CSIT they are in CSC481 Introduction to Cloud Computing, Unit 1 "Introduction to Cloud Computing" (characteristics, advantages, disadvantages) and Unit 4 "Cloud Programming Models" (MapReduce, Hadoop, Dryad, Pig and Hive style parallel models, task and data parallelism). In BIT they are in BIT408 Cloud Computing, Unit 1, Unit 6 "Cloud Programming Models and Parallelization" and Unit 9 "Cloud Availability, Disaster Recovery, and Governance".

This guide gives six advantages you can reproduce in a list, the main disadvantages, availability with the "nines" table, and the programming models with a MapReduce word-count walk-through.

6 advantages of cloud computing

Cloud computing delivers computing resources (servers, storage, databases, software) over the internet on a pay-as-you-go basis. Its usual advantages, in a form that is easy to write down, are these six.

1. Cost saving. There is no up-front purchase of hardware and data centre space. The customer converts capital expenditure into operating expenditure and pays only for what is used.
2. Scalability and elasticity. Resources can be increased during a peak (an admission result day, an exam registration) and released afterwards, automatically if required.
3. Accessibility. Services are available from any place and any device with an internet connection, which supports remote work and collaboration.
4. Reliability and availability. Providers replicate data and run on redundant infrastructure across zones and regions, so a single hardware failure does not stop the service, and backup and disaster recovery are simpler.
5. Speed and agility. A server or a database can be provisioned in minutes through self-service, so projects start quickly and new ideas can be tried cheaply.
6. Maintenance and security handled by the provider. The provider patches, upgrades and monitors the platform and invests in physical and network security that a small organisation could not afford.

Other points that are often added are automatic software updates, resource pooling (multi-tenancy) and measured service. If the question asks for six, give these six with one line of explanation each.

Disadvantages of cloud computing

The balancing list is short. Dependence on the internet: without a connection the service is out of reach. Security and privacy concerns, because data sits on a provider's systems and the customer gives up some control. Vendor lock-in, since moving between providers is costly when services and data formats differ. Limited control and customisation of the underlying infrastructure. Downtime that is outside your control when the provider has an outage. Ongoing cost that can exceed owning hardware for a steady, predictable heavy workload, and compliance questions about where data is stored.

Availability in cloud computing

Availability is the fraction of time a service is up and able to respond correctly. It is normally quoted as a percentage over a year and is the main promise in a Service Level Agreement (SLA).

$$\text{Availability} = \frac{\text{Uptime}}{\text{Uptime} + \text{Downtime}} \times 100%$$

The common way to express it is in "nines". Each extra nine cuts the permitted downtime by a factor of ten.

For example, 99.9% means the service may be down for 0.1% of 365 x 24 = 8760 hours, which is 8.76 hours a year.

Providers raise availability by removing single points of failure. Redundancy means keeping several copies of servers, storage and network paths. Failover means that when a component fails, traffic moves automatically to a healthy standby. Load balancing spreads requests over many instances so that one failure removes only a share of capacity. Multi-zone and multi-region deployment keeps the service alive when a whole data centre or region is lost, and backups with a disaster recovery plan restore it afterwards. The SLA states the promised availability, how it is measured and the credit the customer receives if the provider falls short. Note that the availability of a system built from components in series is the product of their availabilities, so it is lower than the weakest one, while redundant components in parallel raise it.

Cloud programming models

A cloud programming model gives a developer a way to write an application that runs in parallel on a large cluster without handling machine failures, data placement and communication by hand. The models in the syllabus are these.

MapReduce splits a job into a map phase and a reduce phase over key-value pairs. The framework splits the input, runs map tasks on many nodes, groups the results by key (shuffle and sort), runs reduce tasks, and re-executes tasks on a failed node. Hadoop is the open-source implementation, with HDFS for storage.

Dryad, from Microsoft, expresses a job as a directed acyclic graph in which vertices are programs and edges are data channels. It is more general than MapReduce because the graph can have any shape.

Pig and Hive sit on top of Hadoop. Pig offers a scripting language, Pig Latin, for data flows, and Hive offers SQL-like queries (HiveQL). Both are compiled into MapReduce jobs, so analysts need not write Java.

Two kinds of parallelism underlie all of them. In data parallelism the same operation is applied to different parts of the data on different nodes, which is how MapReduce works. In task parallelism different operations or threads run at the same time, possibly on the same data.

MapReduce word-count walk-through

Suppose the input has two lines, split between two map tasks: "cloud is cheap" and "cloud is fast".

The first mapper emits (cloud, 1), (is, 1), (cheap, 1). The second mapper emits (cloud, 1), (is, 1), (fast, 1). During shuffle and sort, the pairs are grouped by key, giving cloud: [1, 1], is: [1, 1], cheap: [1] and fast: [1]. Each reducer sums the list for its keys and writes (cloud, 2), (is, 2), (cheap, 1), (fast, 1).

map(key, line):      for each word w in line: emit(w, 1)
reduce(word, counts): emit(word, sum(counts))

How to write it in the exam

For "list six advantages", give the six numbered points above with a line each; add a two or three item disadvantage list only if asked. For availability, give the definition, the formula, the nines table with at least 99%, 99.9% and 99.99%, and four techniques (redundancy, failover, multi-region, SLA). For programming models, define the term, describe MapReduce with the word-count example and a diagram, mention Hadoop, Dryad, Pig and Hive in one line each, and state data versus task parallelism.

The unit notes are on the Introduction to Cloud Computing notes and the Cloud Programming Models notes. BIT students should read Cloud Availability, Disaster Recovery, and Governance. The subject pages are CSC481 and BIT408.