In this paper, the viability of the throughput and frame loss rate benchmarking procedures of RFC 8219 is tested by executing them to examine the performance of three free software SIIT (also called stateless NAT64) implementations: Jool, TAYGA, and map646. An important methodological problem of the two tested benchmarking procedures is pointed out: they use improper timeout setting. A solution of individually checking the timeout for each frame is proposed to get more reasonable results, and its feasibility is demonstrated. The unreliability of the results caused by the lack of requirement for repeated tests is also pointed out, and the need for relevant number of tests is demonstrated. The possibility of an optional non-zero frame loss acceptance criterion for throughput measurement is also discussed. The benchmarking measurements are performed using two different computer hardware, and all relevant results are disclosed and compared. The performance of the kernel based Jool was found to scale up well with the number of active CPU cores and Jool also significantly outperformed the two other SIIT implementations, which work in the user space.

在本文中，通过执行RFC 8219的吞吐量和帧丢失率基准测试程序来检查三种免费软件SIIT（也称为无状态NAT64）实现的性能来测试它们的可行性：Jool，TAYGA和map646。指出了这两个经过测试的基准程序的重要方法论问题：它们使用了不正确的超时设置。提出了一种针对每个帧分别检查超时的解决方案，以获得更合理的结果，并证明了其可行性。还指出了由于缺乏重复测试的要求而导致的结果不可靠，并且证明了需要进行相应数量的测试。还讨论了用于吞吐量测量的可选非零帧丢失接受准则的可能性。使用两种不同的计算机硬件执行基准测试，并公开和比较所有相关结果。发现基于内核的Jool的性能可以随着活动CPU内核数量的增加而很好地扩展，并且Jool的性能也明显优于其他两个在用户空间中工作的SIIT实现。