zersh01
commented
1 month ago dev_weight = net.core.netdev_budget
The structure of the /proc/net/softnet_stat file is typically as follows (as of the latest Linux kernel versions):
00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
Each field is presented in hexadecimal format. Here's what these fields usually represent:
- processed: The number of processed packets.
- dropped: The number of packets that were dropped due to queue overflow.
- time_squeeze: The number of times netdev_budget was exhausted (i.e., when the number of processed packets reached the limit set by net.core.netdev_budget).
- cpu_collision: Collisions with other CPU processes.
- received_rps: The number of packets processed in the context of RPS (Receive Packet Steering).
- flow_limit_count: The number of times flow limit was exceeded.
The time_squeeze field is particularly important if you are interested in how often the netdev_budget is insufficient for processing all packets in the queue during one softirq call. If this value is high, it may indicate that the current net.core.netdev_budget is too low for the current load and should be increased to improve the performance of the network subsystem.
leandromoreira
daliang111
is there any way to monitor the dev_weight， /proc/net/softnet_stat have no info about dev_weight. like the kernel code, no way to monitor if dev_weight is to small? void __qdisc_run(struct Qdisc q){
int quota = dev_tx_weight; // max send package num int packets;
while (qdisc_restart(q, &packets)) {
/
quota -= packets;
if (quota <= 0 || need_resched()) { //quota is <=0, next softirq and break
__netif_schedule(q);
break;
}
} }