基于BF的单向网络性能抽样测量技术

Technique for One-Way Sampling Measurement of Performance Based on BF

针对在进行网络单向性能测量时,存在抽样效率不高和报文标识均匀性差等问题,提出了一种基于BF(BloomFilter)的单向网络性能抽样测量框架及算法,解决了在不同测量点对上从大量背景流量条件下的抽样有效性和报文标识有效性问题。仿真性能表明,待测报文抽样效率与BF理论分析值偏差为0.5%,且报文标识冲突比例仅为3.5%左右,软件实现抽样时间开销为0.1μs量级;该抽样测量技术十分适合于从大量背景流量中快速抽样出少量待测报文,从而有效地测量出诸如单向传输时延、时延抖动及单向报文丢失率等重要网络性能参数。

The efficiency of current measurement technology is still low when sampling measuring the one-way performance of high-speed network, which is attributed to the limitation of low efficiency of sampling and bad uniformity of packet identification. A new architecture and algorithm for one-way performance measurement based on BF(Bloom Filter) are proposed after detailed research on BF structure and performance of hash functions, which improves the efficiency of sampling and packet identification. Simulations on the architecture and algorithm show that the sampling efficiency is close to theoretical value derived from BF, and the bias between those is 0\^5%; the collision ratio of packet ID is less than 3\^5%; the mean sampling time overhead is on the order of 0\^1 μs. So this architecture and algorithm are feasible for sampling measuring some metrics such as one-way delay, one-way delay variation and one-way packet loss ratio, by only sampling small amount packets out of large amount background traffic with high efficiency.