sometimes we need to find long network latency in heavy traffic product env. and there maybe no packet drop, it's just delay.
like those caused by sk_stream_wait_memory by tcp buffer full, or by TSQ, or by qdisc etc.
we should add option like:
--diag --diag-latency --min-latency 5ms
to catch skb latency from born to death in ingress and egress path.
if --diag-latency option is specified, there will be no output for normal packet, when long latency catched, it will at least show 2 function between the latency spike.
sometimes we need to find long network latency in heavy traffic product env. and there maybe no packet drop, it's just delay. like those caused by
sk_stream_wait_memoryby tcp buffer full, or by TSQ, or by qdisc etc.we should add option like:
to catch skb latency from born to death in ingress and egress path. if --diag-latency option is specified, there will be no output for normal packet, when long latency catched, it will at least show 2 function between the latency spike.