Resource Management I - Week 4

Question 1

Resources in distributed systems

Answer 1

Hardware
Cpu, storage, network links

Software
Web services

Data
- Datasets

Applications need different shares of resources. Distributed systems need to optimise based on the limited number of resources and the applications requests for them.

Question 2

Centralized computing

Answer 2

Primarily used until the 90s. about 200K hosts connected to the internet in 1990.

Data transmission in the order of Kb/s

Still used distribution for compute-intensive applications. E.g. Long Integer Factorisation

The application was written in C, would run on a machine when idle and use email to communicate with a server; sending results of requesting data:
- factorising 100-digit integers would take months, with a good implementation the distributed technique would take just a few days.

Question 3

Grid computing

Answer 3

Grid concept and hardware was too heavyweight

CERN is the main user, data generated in the order of 1PB/s

By 2010 most people talking about cloud computing, GRID didn’t seem to take off

Question 4

Cloud computing

Answer 4

Largely evolved from the grid

Relies on virtualisation of resources that create virtual machines as opposed to physical machines

Idea is on demand resource provisioning - users don’t own the resources they are provided by providers as a service

Question 5

Elasticity

Answer 5

Resources the can be added on demand
Horizontal: add or remove new instances
Vertical: Increase or decrease resource attributes

Question 6

Software as a service (SaaS)

Answer 6

The cloud provider hosts some software which is provided to users.
E.g. Microsoft Office 365, DropBox

Question 7

Platform as a Service (PaaS)

Answer 7

Cloud provider offers an environment that allows the user to create applications (e.g. access to some programming functionality, databases, etc…) E.g: Google Compute Engine

Question 8

Infrastructure as a Service (IaaS)

Answer 8

Cloud provider offers pas-as-you-go access to storage, networking and other resources.

Question 9

Can’t classify everything on the cloud, e.g. Hadoop + mapReduce

Answer 9

Remember this

Question 10

Cloud provider

Answer 10

The owner of resources:
Objectives:
- Maximise income
- Minimise costs (e.g. energy)

Needs to allocate virtual machines to physical machines
- Initial allocation (placement)
- Reallocation

Question 11

Cloud consumer

Answer 11

Objective:
- Obtain resources to complete a task as cheaply (and quickly) as possible

Needs to choose what resource to rent (resource provisioning):
- Overprovisioning: too many resources for the needs of the task
- Underprovisioning: resources are not sufficient for the task

Question 12

Virtual Machine (VM) Allocation

Answer 12

Done by the cloud provider

Many virtual machines (VMs) need to share physical machines (PMs) need to meet objectives:
- Service level agreement with the user (translates to money)
- Maximise resource utilisation
- Minimise energy costs

Needs to do:
- Initial placement and reallocation, when to reallocate?

Mapping VMs to PMs is a more complicated bin packing problem

Question 13

Volunteer Computing

Answer 13

Software that runs in the background of networked resources, downloading tasks and sending the results back. E.g. finding prime numbers.

Concern is it’s not very safe.