Özet:

— In this study; we aimed to propose design of a QoS aware resource management infrastructure for a GPGPU supported Grid computing system. This Grid system consists of hybrid (CPU + CPU) and heterogeneous (Nvidia + AMD Radeon) GPGPU computational nodes. It can manage both small scale unit (connections, threads, buffer pools etc.) and large scale unit (whole computing machines). As increasing of the network communication bandwidth and developing powerful computer hardware (CPU, GPU etc.), distributed computing systems acquire more and more attention day by day. Grid computing is as a major player in such kind of distributed system environments like cloud, volunteer, hybrid and etc. Since it supports large scale resource sharing between geographically distributed computer clusters and even single computers. Nowadays, there is another important technology pillar to implement high performance computing rather than CPU, it is known as GPU computing. The GPU systems are ideal especially to data intensive applications; such as image processing, data mining, financial computations etc. Therefore, GPU based grids give an undertaking higher computational performance. GPU processor consists of lots of controllable cores which can be used for high performance demanded applications. Ultimately, the major concerns in grid computing are particularly related to managing QoS requirements, granularity of resources, and heterogeneous resources (both CPU and GPU).  

Anahtar Kelimeler: