Design and implementation of the high speed TCP/IP Offload Engine

As the growth of Ethernet speed surpassed the growth of microprocessor performance, TCP/IP Offload Engine (TOE) technology has emerged and aimed at not only releasing servers and communication systems from burdened conventional TCP/IP stack but improving the network utilization rate. To lower the risk in developing TOE, one may rely on certain platform containing a single or plurality of embedded microprocessor(s). However, one may find that it is not easy to use such slow-clocked microprocessor(s) to achieve the high speed goal. Hence this paper proposed 5 design guidelines and a corresponding architecture in TOE offloading which come out with experimental results showing that the proposed design further approaches the limit of 1Gbps Ethernet, i.e. 858.82 Mbps as receiving and 793.25 Mbps as transmitting while saves 2/3 to 3/4 computation power compared with using conventional TCP/IP stack.