Tech Primer

High Availability and Cluster Computing

A cluster refers to a group of loosely coupled computers that work together as one. This may be a bunch of workstations connected over a high-speed Local Area Network (LAN) or more likely it may be a bunch of servers tied together using a proprietary or standard high-speed interconnect. Clusters offer greater speed and reliability. Cluster elements usually are inexpensive x86 boxes that run Linux or Windows.

Defining High Availability

High Availability is also referred to as reliability, availability or fault tolerance. High availability is achieved by detecting failures and reconfiguring a computer system by shifting the workload from a failed node or processor to remaining nodes or processors.

The benefits of clustering

Some of the principal benefits of cluster computing are scalability (vertical and horizontal), availability and high performance. Cluster computing is used for running scientific applications, ISPs use it in server farms that run Web and e-commerce software while TV and movie studios use clusters for rendering special effects and animation.

Cluster varieties

Clusters can be configured to offer high availability, high performance or load balancing or any combination thereof.

High availability clusters employ redundant nodes that take over when primary nodes fail. The nodes can be easily removed for maintenance and repair. A non-functioning node can be restored without affecting the availability of the service provided by the cluster. Linux HA or High Availability Linux is a commonly used free software package.

High performance clusters (HPC) on the other hand are used to boost raw performance by splitting a computational task across the nodes in a cluster. This cluster topology is used in scientific computing and, more often than not, you will find these nodes running Linux and free software to implement parallelism.

A load balancing cluster is used to improve performance while maintaining high availability. Such a cluster of computers is known as a server farm. If a node fails, the request is redistributed between the remaining nodes. This architecture is popular in Web hosting environments at ISPs.

The technology

Clustering was developed to permit the use of multitudes of low cost computers to handle tasks that otherwise would require high-end boxes with dozens of processors. Some of the technologies associated with cluster computing are host protocol stack processing and interconnect technologies. MPI or Message Passage Interface, a communication library lets programmers write parallel programs in C and FORTRAN (Formula Translator). Some of the software used in the GNU/Linux world are Beowulf, distcc, LVS (Linux Virtual Server), Mosix, OpenMosix and kerrighed.

For more information visit

www.cs.sandia.gov/cplant
www.linuxhpc.org

—Garima Grover

grover_garima@hotmail.com
For more information visit http://www.top500.org/