On 3/11/20 4:54 PM, Lukas Wunner wrote:
On Wed, Mar 11, 2020 at 03:05:16PM +0100, Daniel Borkmann wrote:
no need to make the fast-path slower for exotic protocols
which can be solved through other means.
As said the fast-path gets faster, not slower.
* Without this commit:
Result: OK: 34240933(c34238375+d2558) usec, 100000000 (60byte,0frags)
2920481pps 1401Mb/sec (1401830880bps) errors: 0
* With this commit:
Result: OK: 33997299(c33994193+d3106) usec, 100000000 (60byte,0frags)
2941410pps 1411Mb/sec (1411876800bps) errors: 0
So you are suggesting that we've just optimized the stack by adding more
hooks to it ...?
Since I've provided numbers to disprove your allegation, I think the
onus is now on you to prove that your allegation holds any water.
Please reproduce the measurements and let's go from there.
This isn't much work, I've made it really easy by providing all the
steps necessary in the commit message.
So in terms of micro-benchmarking with pktgen, if I understand you correctly,
you are basically measuring loopback device by pushing packets through the
__dev_queue_xmit() -> loopback_xmit() -> netif_rx() till the stack drops them
in IP layer on ingress side? I wonder how your perf profile looks like ...
Setting a drop point in tc layer and then measuring the effect before/after
this change with CONFIG_NETFILTER_EGRESS enabled, I'm getting a stable degration
from ~4.123Mpps to ~4.082Mpps with pktgen, definitely not seeing a speedup.
# ip link add dev foo type dummy
# ip link set dev foo up
# tc qdisc add dev foo clsact
# tc filter add dev foo egress bpf da bytecode '1,6 0 0 0,'
# modprobe pktgen
# echo "add_device foo" > /proc/net/pktgen/kpktgend_3
# samples/pktgen/pktgen_bench_xmit_mode_queue_xmit.sh -i 'foo' -n 400000000 -m '11:11:11:11:11:11' -d '1.1.1.1'
Also to let pktgen count the skb:
diff --git a/net/core/pktgen.c b/net/core/pktgen.c
index acc849df60b5..8920da7a7a67 100644
--- a/net/core/pktgen.c
+++ b/net/core/pktgen.c
@@ -3372,11 +3372,11 @@ static void pktgen_xmit(struct pktgen_dev *pkt_dev)
ret = dev_queue_xmit(pkt_dev->skb);
switch (ret) {
case NET_XMIT_SUCCESS:
+ case NET_XMIT_DROP:
pkt_dev->sofar++;
pkt_dev->seq_num++;
pkt_dev->tx_bytes += pkt_dev->last_pkt_size;
break;
- case NET_XMIT_DROP:
case NET_XMIT_CN:
/* These are all valid return codes for a qdisc but
* indicate packets are being dropped or will likely
(In any case, this whole discussion is moot for out of tree code.)
Thanks,
Daniel
