Difference between revisions of "Cluster Computing"
m (→Examples of cluster solutions) |
m (→Project: Mercantec Beowulf) |
||
| Line 43: | Line 43: | ||
The goal of the Mercantec Beowulf are to build a scalable cluster for cracking MD5 hashes or some other similar task which are easily distributed among nodes. <br/> | The goal of the Mercantec Beowulf are to build a scalable cluster for cracking MD5 hashes or some other similar task which are easily distributed among nodes. <br/> | ||
The individual nodes should boot from PXE on the network and automatically enter the cluster. | The individual nodes should boot from PXE on the network and automatically enter the cluster. | ||
| − | [[Category:Network]] | + | [[Category:Network]][[Category:cluster]] |
Revision as of 15:55, 2 April 2009
See Cluster Computing for a definition Engineering a Beowulf-style Compute Cluster
Contents
What is a Cluster
A cluster is a setup of two or more computers - called nodes - that work together to perform a task.
Types of clusters
Basically there are four type of clusters:
| Storage Clusters | Shared filesystem among nodes |
| High Availability Clusters | Failover, redundancy |
| Load Balancing Clusters | Distribute Load between two or more nodes |
| High Performance Clusters | Parallel computing, more nodes solving the same problem |
Often you would need a combination of these Clusters in a installation.
Storage Clusters
High Availability Clusters
Load Balancing Clusters
Load balancing with Cisco IOS NAT
High Performance Clusters
While there are several systems for cluster computing. OSCAR is very popular.
Grid computing
For example seti@home
Examples of cluster solutions
OSCAR
Diskless nodes
See Pegasus homepage for ideas to implement a diskless Linux cluster.
Concepts and abbreviations
Fencing
When a node fails in a cluster, the node is fenced, which means it will be rejected from shared resources.
Linux Clustering at Mercantec
Project: Mercantec Beowulf
The goal of the Mercantec Beowulf are to build a scalable cluster for cracking MD5 hashes or some other similar task which are easily distributed among nodes.
The individual nodes should boot from PXE on the network and automatically enter the cluster.