Cluster Computing & Grid Computing

Cluster Computing

A computer cluster is a group of linked computers, working together closely thus in many respects forming a single computer. The components of a cluster are connected to each other through fast local area networks

J Requirements for computing increasing fast. 

More data to process.

More compute intensive algorithms available.

J  Approaches to supply demand

Qualitative: Optimized algorithms, faster processors, more memory. 

Quantitative: Cluster computing, grid computing, etc.

J Cluster categorizations

High Availability Cluster

Load Balancing Cluster

HPC Cluster

J High Availability Clusters

Failover Clusters, mainly implemented to improve the availability of service that cluster provides 

They operates by having redundant nodes, upon failure the standby node take cares

Types of High availability clusters: one way & two way 

Often used for critical databases, network files sharing and business applications

J Load Balancing Clusters

Multiple computers connected together to share computational workload

Logically they are multiple computers but function as single virtual computer

Request initiated from the user is distributed among all the nodes by one or more load balancer

J HPC Clusters

HPC clusters are mainly used to increases the performance by splitting the computational task into different nodes

Mainly used in scientific computing 

Popular HPC cluster implementations are nodes running with linux os and free software’s to implement the parallelism

The job running on the cluster nodes requires little or no inter nodes communication is called “Grid Computing” 

The local Scheduling software manages the cluster nodes load balancing 

Middleware such as MPI (Message Passing Interface) or PVM (Parallel Virtual Machine) permits compute clustering programs to be portable to a wide variety of clusters

Grid Computing

J What is Grid?

Grid computing is a term referring to the combination of computer resources from multiple administrative domains to reach a common goal.

Coordinates resources that are not subject to centralized control. 

Uses standard, open, general-purpose protocols and interfaces.

Delivers nontrivial qualities of service.

J Why Grid?

Large-scale science and engineering are done through the interaction of people, heterogeneous computing resources, information systems, and instruments, all of which are geographically and organizationally dispersed. 

The overall motivation for “Grids” is to facilitate the routine interactions of these resources in order to support large-scale science and Engineering.

Virtual Organization (VO) refers to a dynamic set of individual and/or institutions defined around a set of resource-sharing rules and conditions 

Multiple organizations that function as one unit through the use of their shared competencies and resources for the purpose of one or more identified goals

Example:  LHC: 1800 Physicists, 150 Institutes, 32 Countries 100 PB of data by 2010; 50,000 CPUs

J Components of Grid

J Grid Architecture

Grid Architecture can be described as the layers of building blocks, where each layer has a specific function, to accomplish Grid Computing Infrastructure

J Grid middleware’s

A mediator layer that provide a consistent and homogeneous access to resources managed locally with different syntax and access method

It provides a uniform interface of the Grid to users and handle all the complexity generated due to heterogeneous systems.

Middleware S/W is a layer between grid applications and low level functionality of grid

J Popular middlware

Globus Toolkit – Globus Alliance 

Glite-EGEE

Gridbus-University of Melbourne

Unicore (Uniform Interface to Computing Resources )-Institutev for Advanced Simulation, Guelich, Germany 

OMII from the Open Middleware Infrastructure Institute

J Functionalities

1.Security

Information Security

Secure communication

Authentication 

Single sign on & Delegation

Authorization

Resource Level 

VO Level

Infrastructure Level Security 

Host Security

2. Job Management

Support an open Job Description Language RSL, JDL, JSDL

Submission, Status Query, Cancel & Destroy, Getting Output & Error

Transferring input/output data from/to remote source/destination

Support Serial/ Parallel Jobs (Heterogeneous & Homogeneous) 

Integration with all Local Resource Managers

3. Data Management

            Two Basic Categories of Data Management

Data Movement

• Secure • Robust • Efficient • Third party movement

Data Replication

• One or more copies or replicas • Survive loss • Easy availability

           Reduce access latency

Performance for distributed applications

4. Information System

Provides mechanism for discovery and monitoring of resources 

Designed to provide various characteristics of resource, computation, service and other entities.

Provide access to static and dynamic information regardingv system components 

Access to information is subject to authentication and authorization mechanisms. 

Information sources are distributed

J Applications

Sequential Jobs for particular platform

Concurrent Sequential Jobs for different platforms

Homogeneous Parallel job for particular OS

Heterogeneous Parallel Jobs

Bio Informatics applications

High Energy Physics Applications

Weather Modelling and Predicting Ocean Currents

Disaster Management

Aerodynamic Simulations

J Advantages

Can solve larger, more complex problems in a shorter time

            Easier to collaborate with other organizations

            Make better use of existing hardware

J Disadvantages 

Grid software and standards are still evolving

 Learning curve to get started

 Non-interactive job submission