Normalization of Heterogeneous Computing Clusters Due to Use of Virtual Nodes

Creation of systems of the distributed calculations is the new perspective tendency of development of computing systems. As uniform definition of a computing cluster does not exist, for simplification of the further statement we shall accept some terminology which will reflect the described properties of a hardware and software platforms essential from the point of view of development of applied programs [1]. The computing cluster is the computing system, consisting of set of the separate computers (nodes) connected among themselves by uniform communication system. One of enough widespread realizations of architecture of the distributed application used for the decision of the big computing tasks is shown on Fig.1. Each node has the local operative memory. Thus the common physical operative memory for sites does not exist. The communication system is used for management and synchronization of work of nodes of a cluster. Usually it allows nodes to cooperate among themselves only by means of messaging. If all nodes included a computing cluster have the same architecture and productivity we deal with a homogeneous computing cluster, otherwise with heterogeneous. Now, when speak about clusters usually mean a homogeneous computing cluster. However to keep a high level of a ratio productivity/cost, it is necessary to use at escalating a cluster the computers most suitable at present which can differ not only on productivity, but also on architecture. Therefore gradually the majority of clusters become heterogeneous clusters. Nevertheless, Any cluster can be considered as the uniform hardware-software system having uniform communication system, center control and planning of loading [2]. The system will consist of the unique client and one or several computing servers working on computers, included nodes of a computing cluster. The basic function of the client is management of work of computing servers. The purpose of the server is the decision of the concrete computing task put by the client.