Skip to Main Content
User-level communications greatly alleviate the software overhead of the communication subsystem by allowing applications to access the network interface directly. Such a direct data path requires the network interface to know the physical memory location of the buffer. Thus, efficient virtual-to-physical address translation is critical. This paper presents a efficient address translation scheme based on address-translation-table where every translation is done on the network interface controller without operating system involvement and miss handling, and zero copy data transfer can be implemented between processes. Adopting this mechanism, we design our communication subsystem oriented to cluster systems based on PCI-Express 2.0. The experimental results show the lowest one-way latency of 2.37us and the peak bandwidth of 6038MB/s, which is the fastest network interface at present.