Metacomputing and Grid Technologies
GRID-technologies are one of the promising implementations in the field of cluster research. The concept of Grid ("grating" - the construction of a computer system like the model of a crystal - with computers at lattice sites) originated in the context of the problem of constructing super-powerful computing devices. In the 80's. XX century. For this, supercomputer technologies have already been applied, and the experience gained was not only positive. It has come to realize that at a high price, the rigid architectures of supercomputers have limited scalability. At the same time, flexible and scalable were required to solve a number of problems (in particular, for forecasting natural phenomena, processing data on high-energy nuclear reactions, for studying the evolution of the universe, etc.) computing power of a fundamentally new level.
The idea of integrating computing and communication technologies into flexible systems was given in the 90s. XX century. push metacomputing as a way to connect individual computing centers. Metacomputing is currently defined as the "use of powerful computing resources transparently accessible through the communication environment". In addition to transparency, features such as seamlessness & quot ;, scalability & quot ;, syncability and globality & quot ;. Thus, it was suggested, "hiding" communication, use computers connected to the network and entire computing centers as a single computing resource. The main emphasis was placed on the fact that a potential user can get almost unlimited capacities for performing supercomplex calculations and data storage .
As processor resources in modern Grid systems, not only powerful specialized computing devices are used, but also workstations and conventional personal computers. Now, under the term computing resources understand everything that participates in the joint computer processing of data. Resources include software packages, communications, storage systems, data warehouses and information systems.
For example, Grid is not just a technology for high-performance computing. Current real content The Grid is an infrastructure that provides support for any globally distributed processing for many types of applications: e-business, distributed production, data mining, High Throughput Computing (HTS) and, of course, distributed supercomputing. Unlike metacomputing, for many applications, including those with a large amount of computation, high-performance communications are not required; in this case, the usual Internet can be used as a communication medium [agora.guru.ru/abrau2008/pdf/050.pdf].
The architecture of the Grid software is briefly described below. The basic Grid software and de-facto international standard is today Globus [osp.ru/os/2003/01/182393/ |. He was taken as a basis in leading Grid projects (IPG, NCSA, Gryphyn, DataGrid). Most of the new research and development in the field of Grid focuses specifically on Globus .
The main task in the Grid is to provide access to resources, and since resources are distributed, the functioning is provided by a special form of software organization - services. Unlike the client-server model this or that set of services is installed here on each resource. A lot of services must satisfy two structural conditions:
• Each type of service must have a standard access protocol, according to which the client application interface is implemented. Within the framework of standard protocols, various ways of implementing services are allowed.
• Multiple services on different resources must be consistent. This presupposes a well-known unification of service sets based on the identity of their semantics, as well as the existence of common rules, regulations and organizational agreements, on which the configuration of services relies.
In Fig. 5.3 shows the algorithm for sending a job and downloading a free computing device, taking into account the access policy and searching for a free resource, implemented by the corresponding service.
Fig. 5.3. Algorithm for passing a job in the Grid system
The success of the Globus project was provided by the following key services and protocols:
• Grid Resource Allocation and Management (GRAM) and Gatekeeper, which provide secure creation and management of remote processes;
• A metadirectory service (Grid Information Service - GIS) responsible for distributed data collection and information services;
• Security infrastructure services (Grid Security Infrastructure - GSI) that support one-time registration, delegation of authority and display of access rights to different local systems.
Works in the field of Grid are aimed today at a significant expansion of the services. The proposed solution at this time is to as a basis Take the Web services standards developed by the W3C consortium (SOAP, WSDL, WS-Inspection, etc.). As a result, the Grid architecture inherits remarkable properties of Web services, while simultaneously inheriting a wide range of development tools and execution environments to choose from (for example, J2EE, Microsoft.Net, IBM WebSphere, Sun ONE). The base architecture of Grid is Open Grid Services Architecture, implemented in Globus 3.0.