Performance is one of the key problems in either high performance computing or GRID application. Performance data must be collected and analyzed for co-allocating resource efficiently, obtaining high performance and f...Performance is one of the key problems in either high performance computing or GRID application. Performance data must be collected and analyzed for co-allocating resource efficiently, obtaining high performance and fault toleration. Furthermore, with the development of Internet and GRID, the exchange of data between virtual organiztions is becoming more and more important, and the type of performance is increasing following the increasing of the resource type, which requires a proper representation of the performance data. This paper does some research on the collection, analysis and representation of the performance data, and presents a Grid-oriented performance tool prototype: THGPT, which can achieve the runtime performance data, describe the data in XML, and implement a browser-based visualization tool of performance data analysis.展开更多
In this article,we have given a elaborate introduction about MVICH implementation underlying MPICH,and illustrated its structure ,working principle and performance. MVICH is an implementation of MPICH ADI2 base ofVIA ...In this article,we have given a elaborate introduction about MVICH implementation underlying MPICH,and illustrated its structure ,working principle and performance. MVICH is an implementation of MPICH ADI2 base ofVIA protocol, using VIA descriptors queue for transferring data in its low layer. It has some characters such as userlevel communications, without kernel trapping, memory binding and registering and so on. MVICH uses buffer(region)management,four transfer protocol,converting dynamically at runtime,and special flow control etc. For highspeed LAN,the combination of VIA and MPI can give a greater improvement for the performance of MPI applica-tions ,and exert those applications' potential ability. It will become the main project of research and development in future.展开更多
SAN(System Area Network)within cluster brings high bandwidths and low latencies to IPC(Inter-Process Communication). However,the communication performance has not been able to makefull use of these hardware improvemen...SAN(System Area Network)within cluster brings high bandwidths and low latencies to IPC(Inter-Process Communication). However,the communication performance has not been able to makefull use of these hardware improvements due to many software overheads. To address this problem,In-tel Corporation,Compaq Computer Corporation, and Microsoft Corporation jointly authored the VIA (Virtual Interface Architecture)specification.As an industrial standard of ULN(User-Level Networking),Ⅵ Architecture defines a set of data operations for message passing, to achieve high-bandwidth, low-latency communication between any couple nodes within cluster ,as well as minimize CPU cycles. VIA gives direct access to network inter-face,bypasses operating system in memory region protected mode ,avoid copies of data during message passing ,and avoid interrupts and context switches if possible.This article describes the principles and mechanism of the Ⅵ Architecture and THVIA.展开更多
文摘Performance is one of the key problems in either high performance computing or GRID application. Performance data must be collected and analyzed for co-allocating resource efficiently, obtaining high performance and fault toleration. Furthermore, with the development of Internet and GRID, the exchange of data between virtual organiztions is becoming more and more important, and the type of performance is increasing following the increasing of the resource type, which requires a proper representation of the performance data. This paper does some research on the collection, analysis and representation of the performance data, and presents a Grid-oriented performance tool prototype: THGPT, which can achieve the runtime performance data, describe the data in XML, and implement a browser-based visualization tool of performance data analysis.
文摘In this article,we have given a elaborate introduction about MVICH implementation underlying MPICH,and illustrated its structure ,working principle and performance. MVICH is an implementation of MPICH ADI2 base ofVIA protocol, using VIA descriptors queue for transferring data in its low layer. It has some characters such as userlevel communications, without kernel trapping, memory binding and registering and so on. MVICH uses buffer(region)management,four transfer protocol,converting dynamically at runtime,and special flow control etc. For highspeed LAN,the combination of VIA and MPI can give a greater improvement for the performance of MPI applica-tions ,and exert those applications' potential ability. It will become the main project of research and development in future.
文摘SAN(System Area Network)within cluster brings high bandwidths and low latencies to IPC(Inter-Process Communication). However,the communication performance has not been able to makefull use of these hardware improvements due to many software overheads. To address this problem,In-tel Corporation,Compaq Computer Corporation, and Microsoft Corporation jointly authored the VIA (Virtual Interface Architecture)specification.As an industrial standard of ULN(User-Level Networking),Ⅵ Architecture defines a set of data operations for message passing, to achieve high-bandwidth, low-latency communication between any couple nodes within cluster ,as well as minimize CPU cycles. VIA gives direct access to network inter-face,bypasses operating system in memory region protected mode ,avoid copies of data during message passing ,and avoid interrupts and context switches if possible.This article describes the principles and mechanism of the Ⅵ Architecture and THVIA.
