Temperature based fault forecasting in computer clusters

摘要

Clusters and Grids have one thing common and that is they both are used to achieve High Performance in Computing. The scope of Cluster is relatively narrow compared to Grid, as Clusters are homogeneous while Grids are heterogeneous. Another emerging area in High Performance Computing (HPC) is Cloud computing that can be considered as a further extension of Grid computing. Apart from other issues that exist in Clusters, Grids and Clouds, there is one common problem or issue that is available in all of them and that is Fault Tolerance and Handling. Fault Tolerance is the technique or the set of techniques that are used when different types of Hardware, Software, Network and other types of problems come during the handling and execution of Clusters, Grids and Clouds. In this research we have focused on fault identification and forecasting from Clusters point of view and have tried to establish a technique that forecasts the faults in Clusters based environments on the basis of temperature. Nodes keep on receiving and monitoring the temperature of the attached devices from temperature sensor and check the temperature threshold values of those devices. If the temperature threshold value of devices is within the range than we place/rate the machine in Green zone. Similarly if temperatures are approaching threshold values then we place the machines in Orange zone that represents that machine may or may not crash on the basis of temperature. Similarly when the devices have crossed the threshold values of the temperature then we place the machine in Red zone that represents that machine is likely to fail due to the failure of one or more hardware devices any time.