Managing network interfaces: Increasing end-point performance in a congested network.

摘要

The history of networking shows us that hardware technology is improving at a rate faster than the networking components that capture and evaluate the traffic are being developed. This is for the most part the result of the inefficiencies in the end-points' software implementations that have been developed for capturing and evaluating the network traffic. Studies show that the hardware supplying the end-points with the packet data vital for performing their operations is progressing at lightning speeds while the algorithms handling the data remain stagnant. By examining the data that is presented to the protocols within the networking stack it can be determined that the congestion that is inherent within a network causes undue stress and delays to the applications that require the data. Latency within the protocols caused by handling traffic that is not destined for them is a problem that exists in all networking protocol implementations.; Existing protocols like TCP have been developed to throttle the sending end-point to control the congestion and provide for a more reliable communications path, but with the increase in traffic in the higher speed networks, the TCP implementations fall short of their goal. Excessive throttling and retries within a specific conversation can cause undue latency and therefore application performance problems and also add to the network congestion problem.; This research provides a solution to the congestion that can be present within the end-point's network stack. By selectively presenting packets to the network stack, congestion can be controlled at the end-point. This provides for a more efficient stack implementation, which results in more packet processing. Efficiency within the packet-processing module provides the basis for more efficient applications that use the data provided them from the networking stack.; A Network Interface Management Algorithm (NIMA) is proposed that allows the networks stack's device driver to reject packets that are not meant for the particular end-point. By rejecting unwanted packets the protocols that use the packets become more efficient. It is likely that most, if not all the accepted packets are useful to the protocols.; Testing the actual throughput that the Gigabit network can handle and the throughput within the end-point shows that the end-point is the bottleneck in the network conversations. A deterministic network can handle 100% of the packets transmitted therefore realizing 100% throughput on the physical layer of the network. The end-point can only handle 2.9% of the throughput making it very unreliable and causing network congestion to occur. Introducing the Network Interface Management Algorithm into the end-points network stack provides a much greater throughput to be realized in the end-point. The Network Interface Management Algorithm provides for a 7.3 percentage point increase over an unmanaged interface greatly improving the performance of the end-point.; This research also provides the development and research communities a look into what can be achieved when the preliminary processing is provided at an earlier stage in the network stack.