Research On Key Technologies Of High-performance Computing Resource Discovery In Multi-domain Environment Based On Semantics

High-performance computing is an important engine that supports scientific discovery and technological innovation,and it has played a huge role in promoting human technological development and social progress.Although the research and development level of China's supercomputer system has been in the forefront of the world,a few highend applications have even reached the international leading level.However,with the development of exascale supercomputers and corresponding applications,the problem of mismatch between the scale of high-performance computing resources and computing service capabilities has become increasingly prominent,making the gap between China's high-performance applications and foreign countries.How to promote the management and use of high-performance computing resources in the current multi-domain environment,improve the service capabilities and resource utilization of highperformance computing infrastructure,and promote the popularity of high-performance computing service capabilities have become the bottleneck problem urgently needed to be researched and solved in this field.In response to these shortcomings,this paper focuses on how to effectively solve the critical and challenging problem of efficient integration and uniform utilization of heterogeneous high-performance computing resources in a multi-domain environment,and carries out research work on key technologies for high-performance computing resource discovery in a semantic multidomain environment.Aiming at providing uniform,efficient and convenient computing services,this dissertation proposes a High-Performance Computing Resource Ontology model named HPCRO.This model integrates information such as resource types,resource attributes,logical relationships,and data transfer among resources and constructs a standardized service specification in corresponding fields.To solve the problems of low discovery efficiency,ontology use thresholds,and high user learning costs existing in the ontology reasoning-based resource discovery methods,this paper proposes a novel resource discovery method to support semantic-based resources fuzzy discovery.Firstly,based on the HPCRO model,we construct a unified high-performance resource pool to realize the integrated identification of heterogeneous software and hardware resources in multiple application domains distributed in a wide area network environment,and provide support for the realization of unified resource management.Secondly,we further propose a Word Net-based quick resource index list based on the pre-reasoning technique.It is valuable to meet the requirements of users for resource discovery and improve the search efficiency of high-performance computing resources.Thirdly,this paper implements the WQRIL resource discovery algorithm which extends hypernym and hyponym semantic relationships of ontology concepts and considers the semantic features of both Word Net and ontology model comprehensively.The experimental results show that our method can achieve better results than others in terms of execution efficiency and resource discovery quality,,which can effectively improve high-performance resource utilization.To overcome the problems single service mode of high-performance computing center,high threshold of resource usage,and low ease of use of resource discovery methods,this paper proposed a joint semantic similarity with ontology model method for high-performance resource discovery.The method optimizes the existing semantic path calculation method by combining the concept depth and local density information contained in Word Net,and proposes a new shortest path calculation method based on IC weighting,which makes the path calculation strategy more in line with judgment criteria of domain experts.Then we further design and implement a semantic similarity calculation method based on the IC-weighted shortest path,fully consider the impact of different semantic attribute information on the semantic similarity calculation.In view of the problem that Word Net cannot provide comprehensive semantic information for all concepts,this paper uses the word vector semantic similarity calculation method to perform semantic enhancement and optimize the calculation results.Finally,the optimized semantic similarity calculation method is integrated with the high-performance computing resource ontology model to find the ontology concept closest to the user's needs through semantic similarity calculation,and then use the index list to provide users with resource matching results.Experimental results show that this method can effectively improve the quality of resource discovery and reduce the difficulty of using resource discovery methods.