Research On Lightweight And Fine-grained Microburst Measurement Method

Microburst is a new type of network anomaly in highspeed networks.Microbursts cause instantaneous packet loss,which seriously affects network performance.At the same time,attacks such as port scanning may also induce microbursts.Detecting and extracting microburst flows is of great significance for troubleshooting network faults,ensuring network stability,and satisfying application quality of service.and the coarse-grained network management system in the traditional network cannot even detect the microburst.The fine-grained port mirroring measurement method clones all packets,which will result in a huge bandwidth overhead.Aiming at the problem that the current measurement scheme cannot take into account the advantages of fine-grained detection of microburst and low resource overhead,a lightweight and fine-grained microburst measurement method based on the sketch data structure is proposed.In order to detect microbursts at a fine-grained level,the features of programmable switch architecture is used to measure the queuing delay of each packet,filter out microburst packets and classify them into medium and high congestion flows.In order to reduce resource overhead,the switch mirrors the packet header after removing the payload of the medium-congestion packets,and uses the interleaved sketch data structure to compress and encode the high-congestion flow information.Only medium congestion packet header and the content of compact sketch data structure is transmitted to the controller,without mirroring and transmitting all original microburst packets,which greatly reduces the bandwidth overhead of mirroring.On the P4 hardware switch equipped with Tofino chip,software simulation and hardware traffic replay experiments are performed using open source packet traces.The experimental results show that the recall rate of microburst flow is greater than 99.5%,the size error of microburst flow is less than 0.06%,and the bandwidth overhead is reduced to less than 1/500 of the in-band telemetry scheme,which is both fine-grained and lightweight.In the future,the microburst flow information storage algorithm can be further optimized,and attempts are made to alleviate the micro-burst phenomenon on the data plane.