© Systems Research Group, Department of Computer Science, The University of Hong Kong.

Main ** People ** Research ** Publications ** Facilities ** Links

Directed Point (DP) High Speed Communication Interface

Recently, the microprocessor clock speeds have approached gigahertz range. Such explosive growth of processor performance has greatly improved the computation. Nevertheless, it stresses the need of a higher speed communication subsystem that can achieve low-overhead calls to access the interconnect, as the performance of a parallel application depends largely on the performance of the interconnection network and the communication software. This paper studies the practical issues on the design of a new communication subsystem, named Directed Point, on a server cluster with Gigabit Ethernet connection, with the goals of achieving high performance and good programmability. The design of Directed Point exploits the underlying network architecture and operating system characteristics to effectively utilize the network and cluster node resources. Several optimizing techniques are proposed to achieve low-latency and high-bandwidth communication. Testing shows that our Directed Point communication subsystem can achieve a 16.3 microseconds single-trip latency and 79.5 MB/s bandwidth on a Gigabit Ethernet network. A realistic communication model is used to calibrate the performance results and to assess various design tradeoffs. To achieve good programmability, we proposed an abstraction model that allows all inter-process communication patterns to be easily coded using the provided API without long learning period. The API of DP preserves the syntax and semantics of traditional UNIX I/O operations, making it easy to use.

Please click here for more details on the Directed Point project, with the source code download and user guide.

Back to Research Main page.

Main ** People ** Research ** Publications ** Facilities ** Links