دانلود رایگان مقاله انگلیسی یک نظرسنجی در مورد سیستم های چندگانه موازی و توزیع شده برای شبیه سازی محاسبات با کارایی بالا به همراه ترجمه فارسی
عنوان فارسی مقاله: | یک نظرسنجی در مورد سیستم های چندگانه موازی و توزیع شده برای شبیه سازی محاسبات با کارایی بالا |
عنوان انگلیسی مقاله: | A survey on parallel and distributed multi-agent systems for high performance computing simulations |
رشته های مرتبط: | مهندسی کامپیوتر و مهندسی فناوری اطلاعات، معماری سیستم های کامپیوتری و سخت افزار کامپیوتر، شبکه های کامپیوتری و سیستم های چندرسانه ای |
فرمت مقالات رایگان | مقالات انگلیسی و ترجمه های فارسی رایگان با فرمت PDF میباشند |
کیفیت ترجمه | کیفیت ترجمه این مقاله خوب میباشد |
توضیحات | ترجمه نصف این مقاله انجام شده است. |
نشریه | الزویر – Elsevier |
کد محصول | f383 |
مقاله انگلیسی رایگان |
دانلود رایگان مقاله انگلیسی |
ترجمه فارسی رایگان |
دانلود رایگان ترجمه مقاله |
جستجوی ترجمه مقالات | جستجوی ترجمه مقالات مهندسی کامپیوتر |
بخشی از ترجمه فارسی مقاله: 6.3.1 D – MASON پروجکشن شبکه : پروجکشن شبکه [28] روشی برای نمایش گراف ساختاری است . (شکل 5 ) نموداری از پروجکشن شبکه را با دو عامل پراکندگی در دو پردازنده نشان می دهد . به منظور تقسیم گراف در چند پردازنده در حالیکه ارتباط بین راس ها در پردازنده های مختلف حفظ میشود . یک کپی از لبه هاو راس های مجاور ساخته شده اند . متاسفانه هیچ اطلاتی در مورد RepastHPC که چطور گراف ها در چند پردازنده که به ندرت استفاده می شود موجود نیست. در مورد D-MASON که برای چند پروجکشن در مدل های یکسان استفاده می شود . به عنوان مثال ، یک پروجکشن شبکه ایی برای یک گرافیک محیطی و یک پروجکشن شبکه ایی برای تعامل بین عوامل است . |
بخشی از مقاله انگلیسی: 6.3.1. D-MASON From the distribution view point, D-MASON is a generic platform as agents could have or could not have a position in a Cartesian space representing the environment. The environment is divided into cells, or partitions, that are assigned to processors to distribute the simulation. D-MASON proposes three ways to distribute a simulation over several cores. Two ways are based on field partitioning or grid partitioning and the last one is based on network partitioning. Overlapping areas (Area of Interest or AOI) are defined to guarantee the continuity of agent perception through node borders. Field or Grid partitioning. Fig. 2 represents the two distribution mechanisms available in D-MASON for grid partitioning. The “Y Distribution” consists in dividing the environment in horizontal cells on the Y axis as shown in Fig. 2(a). The “XY Distribution” consists in dividing the environment in square cells using the X and Y axes as shown in Fig. 2(b). The grid environment is distributed by cutting its cells and mapping them on the nodes of the execution platform. To provide a continuous environment, D-MASON uses overlapping zones, also called “Area of Interest” (AOI) as shown in Fig. 3. This mechanism consists in locally copying a part of each neighbor cell or adjacent partition. The overlapping zone offers the opportunity for agents to keep a global perception field even when their environment is cut between several nodes. To maintain the continuity each cell exchanges information with its direct neighbors before each time step. During the distribution phase only the environment is considered for dividing the simulation. Agents are not takeninto account for this phase. Agents density is however taken into account for the load balancing phase. Network partitioning. Network field is a way to represent a graph structure in the simulation. To distribute a graph over several processors, frameworks like ParMetis [51] or Metis are used. ParMetis is an MPI based parallel framework that implements algorithms for partitioning unstructured graphs, meshes and so on. ParMetis extends the functionality of Metis to distributed and large scaled graphs. We did not find more information about the network partitioning mechanism which was added in a recent update of the platform. It is important to note that, in the D-MASON platform, we can use different types of layer in the same simulation. In other terms, a spatial distribution could be used for the environment while a network partitioning is used to represent interaction between agents. In that case, the distribution is only based on the field partitioning. To distribute agent simulations, RepastHPC uses a mechanism called “Projection”, adapted from the Repast S platform. It represents the environment in which the agents evolve. Projections can be of two types, either grid or network. The projections are used to impose a structure to the agents in which they can evolve. Overlapping areas (or area of recovery) are defined to manage the continuity of agent perception through node borders. 6.3.2. RepastHPC Distribution in RepastHPC as several similarities with the DMASON proposition: grid projections are roughly equivalent to grid partitioning and a network projection is equivalent to a network partitioning. Projection Grid. The Grid Projection represents agents in a Cartesian space. To distribute a model over several processors, the environment is divided into cells of equal size. These cells are regularly distributed on the processors. Each processor is responsible for a sub-part of the grid. The sub-parts of the grid are connected with overlapping areas. Fig. 4 represents a distribution schema of a grid projection on 4 processors. The grid ranges from coordinate (0,0) to coordinate (5,7). Processor P1 is responsible for the sub-part (0.0) × (2.3), P2 is responsible for the portion (3.0) × (5.3) and so on. In our example, the size of the overlapping areas is set to 1. In this case, P1 contains an area buffer that includes the entire column 3 of processor P2 and line 4 of processor P3. Network Projection. Network Projection [28] is a way to represent a graph structure. Fig. 5 represents a diagram of a network projection with two agents distributed on 2 processors. In order to divide the graph on several processors while maintaining the links between vertices dispatched on different processors, a copy of neighbor edges and vertices is made. Unfortunately, no information is available in the RepastHPC documentation on how the graph is distributed on multiple processors so that this projection is hardly usable. As for D-MASON it is possible to use several projections for the same model. For instance, a Grid projection for a geographical environment and a Network projection for agent interactions. |