PROCESSING OF DATA NETWORK STREAMS IN RECONFIGURABLE COMPUTING SYSTEMS

Abstract

In the paper the author considers applicability of a “pipeline-in-pipeline” method for the tasks of network data flow real-time processing. Owing to the method, it is possible to minimize hardware costs, if recursive algorithms are used: for example, for tasks of encryption or reduc-tion of network traffic. In this case, new data cannot be processed, until the result of previous data processing is received. The classical solution implies concurrent use of several devices, each of which processes its own dataflow independently. The “pipeline-in-pipeline” method allows using a single device for processing, which receives independent data from concurrent flows. However, such solution is efficient only for network packets of the same size. We have suggested a modification of the method, which can provide real-time processing of different size data packets. Network data transfer protocols require keeping of packets sequence order. A standard solution implies use of a dispatcher, which optimizes filling of input data buffers and additional buffers after the processing device, synchronizing the packets. For reduction of hardware costs, we have suggested an alternative solution, which implies use of the same memory for initial and processed data. Network packets are sent to special blocks, which areconnected sequentially and which consist of dual-port memory and a unit of data flows and memory state control. We also suggest using of reserve units to exclude the situation of memory overflow. We have estimated the minimum hardware costs of FPGA for the tasks of network packets processing with the help of recursive algorithms at the rate of 10 Gb/sec. Analysis of results has proved efficiency of the suggested method in comparison with existing solutions. So, it is possible to increase the maximum data processing rate in more than 2.5 times for the speci-fied hardware recourse. We recommend to use the suggested modification of the “pipeline-in-pipeline” method for tasks of data encryption and reduction in high-speed computer networks.