This paper presents a new scalable interconnection topology called the Leafy Tor Cube (LTC) for high end computing system. As compared to the other interconnection networks LTC is found to be more attractive in terms of topological attributes like degree connectivity, edges, average node distance and message traffic density etc. It helps in improving the node packing density for high performance computing. Our proposed network is highly scalable and also robust in nature. The different types of performance metrics show that the LTC is a better candidate for distributed processing and high performance computing than its predecessors. As the number of nodes of the proposed network is sufficiently more so this network is a preferable one for high performance computing (HPC) with larger volume of communication. Also it is shown that LTC network is more scalable than other important topologies. With increase in node count, the message density for LTC is less compared to some other popular topologies like hyper cube(HC), crossed cube(CC), leafy cube(LC). Less message traffic density also proves that the LTC is an efficient network when traffic congestion is considered. The degree of LTC is more because of child nodes which can be a preferable backup in high performance computing scenario. This property shows that LTC network is more tolerant to fault and reliable than LC. LTC has decreased diameter as compared to LC. The routing and broadcasting algorithms for LTC are proposed with lesser time complexity.

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

Leafy Tor-Cube: A New Scalable Hybrid Interconnection Network for High Performance Computers

  • Rashmita Padhi,
  • Nibedita Adhikari

摘要

This paper presents a new scalable interconnection topology called the Leafy Tor Cube (LTC) for high end computing system. As compared to the other interconnection networks LTC is found to be more attractive in terms of topological attributes like degree connectivity, edges, average node distance and message traffic density etc. It helps in improving the node packing density for high performance computing. Our proposed network is highly scalable and also robust in nature. The different types of performance metrics show that the LTC is a better candidate for distributed processing and high performance computing than its predecessors. As the number of nodes of the proposed network is sufficiently more so this network is a preferable one for high performance computing (HPC) with larger volume of communication. Also it is shown that LTC network is more scalable than other important topologies. With increase in node count, the message density for LTC is less compared to some other popular topologies like hyper cube(HC), crossed cube(CC), leafy cube(LC). Less message traffic density also proves that the LTC is an efficient network when traffic congestion is considered. The degree of LTC is more because of child nodes which can be a preferable backup in high performance computing scenario. This property shows that LTC network is more tolerant to fault and reliable than LC. LTC has decreased diameter as compared to LC. The routing and broadcasting algorithms for LTC are proposed with lesser time complexity.