Re: [PATCH mptcp-next v7 16/19] selftests/bpf: Add bpf_burst test

[Geliang Tang <geliang.tang@suse.com>]

Hi Paolo,

On Wed, Jun 07, 2023 at 07:08:59PM +0200, Paolo Abeni wrote:
> On Wed, 2023-06-07 at 16:38 +0800, Geliang Tang wrote:
> > This patch adds the burst BPF MPTCP scheduler test: test_burst(). Use
> > sysctl to set net.mptcp.scheduler to use this sched. Add two veth net
> > devices to simulate the multiple addresses case. Use 'ip mptcp endpoint'
> > command to add the new endpoint ADDR_2 to PM netlink. Send data and check
> > bytes_sent of 'ss' output after it to make sure the data has been sent
> > on both net devices.
> > 
> > Signed-off-by: Geliang Tang <geliang.tang@suse.com>
> > ---
> >  .../testing/selftests/bpf/prog_tests/mptcp.c  | 38 +++++++++++++++++++
> >  1 file changed, 38 insertions(+)
> > 
> > diff --git a/tools/testing/selftests/bpf/prog_tests/mptcp.c b/tools/testing/selftests/bpf/prog_tests/mptcp.c
> > index a968641cc94a..b9f6dcf995fd 100644
> > --- a/tools/testing/selftests/bpf/prog_tests/mptcp.c
> > +++ b/tools/testing/selftests/bpf/prog_tests/mptcp.c
> > @@ -10,6 +10,7 @@
> >  #include "mptcp_bpf_bkup.skel.h"
> >  #include "mptcp_bpf_rr.skel.h"
> >  #include "mptcp_bpf_red.skel.h"
> > +#include "mptcp_bpf_burst.skel.h"
> >  
> >  char NS_TEST[32];
> >  
> > @@ -455,6 +456,41 @@ static void test_red(void)
> >  	mptcp_bpf_red__destroy(red_skel);
> >  }
> >  
> > +static void test_burst(void)
> > +{
> > +	struct mptcp_bpf_burst *burst_skel;
> > +	int server_fd, client_fd;
> > +	struct nstoken *nstoken;
> > +	struct bpf_link *link;
> > +
> > +	burst_skel = mptcp_bpf_burst__open_and_load();
> > +	if (!ASSERT_OK_PTR(burst_skel, "bpf_burst__open_and_load"))
> > +		return;
> > +
> > +	link = bpf_map__attach_struct_ops(burst_skel->maps.burst);
> > +	if (!ASSERT_OK_PTR(link, "bpf_map__attach_struct_ops")) {
> > +		mptcp_bpf_burst__destroy(burst_skel);
> > +		return;
> > +	}
> > +
> > +	nstoken = sched_init("subflow", "bpf_burst");
> > +	if (!ASSERT_OK_PTR(nstoken, "sched_init:bpf_burst"))
> > +		goto fail;
> > +	server_fd = start_mptcp_server(AF_INET, ADDR_1, 0, 0);
> > +	client_fd = connect_to_fd(server_fd, 0);
> > +
> > +	send_data(server_fd, client_fd);
> > +	ASSERT_OK(has_bytes_sent(ADDR_1), "has_bytes_sent addr 1");
> > +	ASSERT_OK(has_bytes_sent(ADDR_2), "has_bytes_sent addr 2");
> > +
> > +	close(client_fd);
> > +	close(server_fd);
> > +fail:
> > +	cleanup_netns(nstoken);
> > +	bpf_link__destroy(link);
> > +	mptcp_bpf_burst__destroy(burst_skel);
> > +}
> > +
> >  void test_mptcp(void)
> >  {
> >  	if (test__start_subtest("base"))
> > @@ -467,4 +503,6 @@ void test_mptcp(void)
> >  		test_rr();
> >  	if (test__start_subtest("red"))
> >  		test_red();
> > +	if (test__start_subtest("burst"))
> > +		test_burst();
> >  }
> 
> I would be curious to see how the ebpf scheduler compare with the
> build-in one performance wise (max tput on 2 fast links with a subflow
> on each of them). The link should be fast enough so that the tput will
> be CPU bound.

I just sent a patch in ML to add time metrics for BPF tests to compare
the performance of each schedulers, named "selftests: bpf: Add time
metrics for sched tests".

> sudo ./test_progs -t mptcp -v | grep " ms"  
default: 178 ms
bpf_first: 6 ms
bpf_bkup: 6 ms
bpf_rr: 24 ms
bpf_red: 9 ms
bpf_burst: 179 ms
bpf_stale: 8 ms

From this result, it can be seen that bpf_burst and default in-kernel
scheduler performance are basically the same.

Thanks,
-Geliang

> 
> /P
>