Wednesday, 24 August 2011

Cluster Supercomputer

Cluster Supercomputer

The addition of nodes to build a cluster to support file sharing, functions, applications and websites all point to one thing – building a supercomputer. The collection of nodes that would eventually form a cluster would be part of something bigger which could easily support almost any type of workload without showing any signs of downtime.

The idea of preventing single point of failure, adding monitoring and measurement tools and even considering the external factors that might jeopardize infrastructure are also small parts towards the ultimate goal of building a supercomputer that will not post any problems while computing.

Hardware Considerations

The first thing administrators or developers would consider in building a supercomputer for varied interests is to have the right hardware.

Processors – building a supercomputer is just including a dual-core processor so that it could provide better processing speed. There are companies who have custom designed nodes wherein the boards can include two or more processors which could run in full speed at the same time.

Inclusion of GPU – although this was first thought to be used for gaming purposes, GPU (Graphics Processing Unit) are now being introduced to boost the processing speed. With the addition of right connectors, GPU could easily improve the performance of the cluster. A good example is when Iraq was suspected to be building supercomputers by buying PlayStation 2.

Memory - Processors and GPU would still need the help of an efficient memory (RAM) if they wanted to use the cluster for application processing.

Applications for Supercomputer

Even though you have the best hardware for a supercomputer through clustering, you will still need to have the right application to ensure the clusters are working together.

OS (Operating System) – The operating system suggests the environment of the cluster and their configuration will suggest how the nodes will interact with each other. It doesn’t matter what OS the user opts to use in accessing files or requesting functions. What is important is that the OS is flexible enough to adapt to different environments and user demands. Because of the needed flexibility and stability, the most popular OS today for clustering is powered by Linux and UNIX.

Applications – Supporting the operating system is the applications that will enable communication and interaction of nodes. Applications based on MPI are currently one of the most recognized set of applications that could easily implement the specific administrator commands. MPI (Message Passing Interface) could be used by programming developers to build their clustering application so that it would fit based on their need. Microsoft, Open Source and Linux also have separate offerings but they still include some libraries based on MPI.

Tools – aside from applications that could run the clusters, additional tools should be integrated to ensure continuous operation. Measurement and monitoring tools are very important to be integrated in clusters and nodes since they provide the much needed timely or even real time feedback to administrators in case one of the nodes fails to perform per expectation.

Keeping up with the Times

The biggest challenge any business would face when they are opting to build a highly efficient clustering system is the evolving technology. Hardware that is deemed powerful today might not yield the same efficiency two years from today.

This is actually not the fault of the business or the hardware but technology whether hardware and software, is rapidly evolving. This would mean constant funding from the business if they wanted to keep up with the latest in technology. Proper planning is also required to avoid too much spending on upgrades.

User Specific

Businesses don’t just develop a powerful clustering system for their use. Like any business plans, clustering has to be done with careful planning and even the possibility of ROI (return on investment) on the part of users. The amount spent in developing clusters and improving nodes should have at least an indirect impact on the improvement of the user experience.

Without the end user in mind, excessive spending on upgrades could happen as the upgrades may not be applicable because of incompatibility. Clustering should be geared towards improvement of user experience which will greatly benefit the company as they offer ease of transaction and improved customer experience.

No comments:

Post a Comment