In fact, doing taskset only on the sending side is sufficient.
事实上，在发送端仅使用taskset就足够了。
"Taskset 0, 2" on the sending side seems to be a clear winner (1.1e9 versus 7.5e8).
发送端的“taskset0,2 ”似乎是全胜的赢家(1.1e9与7.5e8)。
Without the taskset, the threads of the sender migrate from one processor to another on a regular basis.
没有taskset，发送方的线程会定期从一个处理器迁移到另一个处理器。
Further runs with just the taskset and just the CHRT indicate that usually only the taskset is needed for these results.
进一步运行taskset和chrt表明，通常这些结果仅需要taskset。
This seems to be because they tend to end up on different chips, which a "taskset 0, 3" confirms as being bad for performance.
出现这种情况的原因是因为它们在不同的芯片上运行，“taskset0,3 ”会得到较差的性能。
The "usually" is due to a strange set of observations I've made, where the throughput was poor when only the taskset was done.
“通常”是基于我已进行的一组观察，当仅有taskset完成时吞吐量会很少。
They suggest that taskset be used to control placement of the threads of the application, and that the CHRT command be used to change the scheduling.
他们推荐了用于控制应用程序线程位置的taskset，以及用于更改调度的chrt命令。
For the taskset-chrt measurement set added in Figure 2, all of the threads of each process are set to CPU 0 and 2 (taskset) only, with SCHED_RR and a priority of 99 (CHRT).
对于在图2中添加的taskset -chrt测量集合，每个进程的所有线程都仅设置为cpu0和2 (taskset)，并具有SCHED _ RR和优先级99 (chrt)。
For the taskset-chrt measurement set added in Figure 2, all of the threads of each process are set to CPU 0 and 2 (taskset) only, with SCHED_RR and a priority of 99 (CHRT).
对于在图2中添加的taskset -chrt测量集合，每个进程的所有线程都仅设置为cpu0和2 (taskset)，并具有SCHED _ RR和优先级99 (chrt)。
应用推荐