Computer clusters

The large computing power availability is a basic factor in many scientific research sectors and it is assuming a more and more important role also in many applicative areas as industrial design, large databases and recent Internet services.

For computing power requirements of various types of applications there is nowadays an interesting alternative to very expensive parallel supercomputers: computer clusters. These architectures provide computing power in a flexible, scalable and especially very low-cost way.

What is parallel programming?

By means of parallel programming you can split a program in multiple tasks running on different processors which exchange data in cooperation. So you can take advantage of memory and calculation power of many computers in parallel, while considerably decreasing the time of programs execution.

To know more... (Introduction to Parallel Computing, from LLNL web site)

What are computer clusters?

A cluster is made up of a group of personal computers, servers or workstatiuons interconnected by a fast network. The cluster nodes have neither monitor, nor keyboard, but they have powerful processors and lots of RAM memory. Then Linux operating system and some standard tools for parallel programming (as for example the MPI library, also used in supercomputers) are installed in all the computers of the cluster.
The computer clusters main advantages compared to commercial supercomputers and workstations are the following:

• Low cost:
  from 5 to 10-15 times less expensive compared to commercial workstations and supercomputers.
• Scalability:
  clusters have a very flexible architecture, you can build systems ranging from few computers (3 or 4) up to many tens of units (64 - 128) with an almost proportional cost.
• Ease of upgrading and maintenance:
  a cluster can be progressively updated by repairing or upgrading its components with commodity hardware, easily available on PC market.
• Standard parallel platform:
  parallel programming software and libraries available for clusters are the same present on commercial supercomputers and workstations (ex. MPI ). So the programs can be compiled on both platforms without any change.
• Open Source software:
  the most of the software installed on clusters is open-source, freely available, extremely tested and reliable, and continuously updated.

Speed and power for scientific computing

Numerical simulations typically require significant computation and memory resources, and often, in spite of technology evolution, only one computer may be not enough. Then it is necessary to simplify the simulation model in different ways, depending on the limitation of the available computational resources.
Computer clusters and parallel programming can meet these needs, speeding up those heavy simulations which require too much time on a single computer. Moreover they allow a very powerful, flexible and scalable architecture, able to adapt its computation and memory capability to each application requirement.

Clusters: a question of costs

Computer clusters are a rather recent technology which purpose is to apply supercomputer solutions to common hardware, saving a lot of money. The task for the parallel machine is divided among the cluster nodes and the communication among processors is performed through a local network.
Extreme cheapness of hardware components: indeed it is possible to buy a cluster with higher performance than a workstation, paying less money. Then a cluster is very scalable, and you can build it ranging from few computers up to hundreds of nodes, obtaining a machine comparable with medium-low parallel supercomputers.
Supercomputers or workstations performances being equal, computer clusters can reach price ratios up to 10-15 times lower.

The power of cooperation

The stregth of clusters is due to the collaboration of many computing units in order to solve a complex problem, by dividing it in many tasks appropriately assigned to each processor. To make the simulation consistent, the cluster nodes can exchange data during the computation. These features highlight how clusters are based on an intelligent collaboration philosophy, which makes a remarkable power for scientific computations available.

Scientific computing applications

In the following there are some applications where computer clusters and parallel programming are used in scientific fields:

• Computational chemistry
• Molecular biology
• High energy physics
• Meteorology
• Astrophysics
• Fluidodynamics
• Material science
• Economy

Contact us to have more information on the use of parallel computing in a particular field.

Back