On Mon, Sep 5, 2022 at 4:31 PM Eelco Chaudron wrote:
>
>
> On 5 Sep 2022, at 12:45, Ales Musil wrote:
>
> > On Fri, Sep 2, 2022 at 4:19 PM Eelco Chaudron
> wrote:
> >
> >>
> >>
> >> On 9 Aug 2022, at 9:17, Ales Musil wrote:
> >>
> >>> When the packet was traveling through patch port boundary
>
Hi guys，
I meet a question about ovs dpdk multiple queues.
When testing tcp band width with iperf3 or qperf ,all packets are only
in one rxq.And the bw result is too low.
Would you please help me?
Version
ovs-vsctl (Open vSwitch) 2.11.0
DPDK 18.11.9
When using a userspace vport ("vxlan0"), dpif-netdev adds an additional
netdev ("vxlan_sys_4789"). The dpif netdev ("vxlan0") is added to the
netdev-offload ports map, thus flows are associated on this netdev.
However, flushing is done on the dpif-netdev level ("vxlan_sys_4789"),
and relevant
Vport's offloads are done on the tracked orig-in-port, but the flow itself
is associated in the vport's map.
Removing the physdev will flush all the ports that are on its map, but
not the ones on other netdevs' maps. Since flows take reference count on
both their vport and their physdev, the
On 5 Sep 2022, at 12:45, Ales Musil wrote:
> On Fri, Sep 2, 2022 at 4:19 PM Eelco Chaudron wrote:
>
>>
>>
>> On 9 Aug 2022, at 9:17, Ales Musil wrote:
>>
>>> When the packet was traveling through patch port boundary
>>> OvS would check if any of the actions is reversible,
>>> if not it would
The goal was to run 200 successful tc tests in a row. To do this the
following was run:
for i in {1..200}; do make check-offloads TESTSUITEFLAGS="50" && break; \
echo "ALL_200_OK: $i"; done;
Unfortunately, a bunch of test cases showed occasional failures.
For now, they are excluded from
With some datapaths, read TC, it takes a bit longer to update the
OpenFlow statistics. Rather than adding an additional delay, try
to read the counters multiple times until we get the desired value.
Signed-off-by: Eelco Chaudron
---
tests/system-offloads-testsuite.at |2 --
tc does not include ethernet header length in packet byte count.
This fix will allow the packets that go trough tc to be 14 bytes less.
This difference in the TC implementation is already described in
tc-offload.rst.
Signed-off-by: Eelco Chaudron
---
tests/system-offloads-testsuite.at |1 -
The tc conntrack implementation does not support the timeout option.
The current implementation is silently ignoring the timeout option
by adding a general conntrack entry.
This patch will skip the related test by overriding the support macro.
Signed-off-by: Eelco Chaudron
---
If a tc flow was installed but has not yet been used, report it as such.
In addition, add a delay to the "IGMP - flood under normal action" test
case to make it work with many repetitions. This delay is also present
in other ICMP/IGMP tests.
Signed-off-by: Eelco Chaudron
---
lib/tc.c
Make the order of the Netlink attributes for odp_flow_key_from_flow__()
the same as the kernel will return them.
This will make sure the attributes displayed in the dpctl/dump-flows
output appear in the same order for all datapath.
Signed-off-by: Eelco Chaudron
---
lib/odp-util.c
tc does not support conntrack ALGs. Even worse, with tc enabled, they
should not be used/configured at all. This is because even though TC
will ignore the rules with ALG configured, i.e., they will flow through
the kernel module, return traffic might flow through a tc conntrack
rule, and it will
Flush datapath flows as TC flows take some more time to be flushed out.
The flush speeds this up.
Signed-off-by: Eelco Chaudron
---
tests/system-offloads-testsuite.at |2 --
tests/system-traffic.at|6 ++
2 files changed, 6 insertions(+), 2 deletions(-)
diff --git
This change fixes the "conntrack - floating" test for the TC
offload case. In this scenario, the connection might move to
CLOSE_WAIT, which would fail the test as it only accepts
TIME_WAIT. However, both indicate the connection was
established, so the test should pass.
Signed-off-by: Eelco
This patch adds a delay before dumping the conntrack table because with
tc it takes a bit longer before it gets synced.
Signed-off-by: Eelco Chaudron
---
tests/system-common-macros.at |3 +
tests/system-offloads-testsuite.at | 25 +
tests/system-traffic.at| 198
tc was not setting the OVS_CT_ATTR_FORCE_COMMIT flag when a forced
commit was requested. This patch will fix this.
Signed-off-by: Eelco Chaudron
---
lib/netdev-offload-tc.c| 13 +++--
tests/system-offloads-testsuite.at |1 -
2 files changed, 11 insertions(+), 3
The revalidator/purge commands in the system test cases sometimes
get called immediately after a partial test is completed. This
could cause the revalidator thread to log an error that it can
not find/delete a flow due to the slower flow installation nature
of TC.
This patch uses a macro to call
TC flower does not allow the push of the implicit null labels (RFC3032).
Avoid the use of such labels in the MPLS test cases.
Signed-off-by: Eelco Chaudron
---
tests/system-offloads-testsuite.at |2 --
tests/system-traffic.at|8
2 files changed, 4 insertions(+), 6
Include and run the system-traffic.at tests as part of the system offload
testsuite. Exclude all the tests that will not run without any special
modifications.
Signed-off-by: Eelco Chaudron
---
tests/system-offloads-testsuite.at | 93
1 file changed, 93
When the test description is part of the OVS_TEST_SKIP_LIST
variable, the test is skipped.
Signed-off-by: Eelco Chaudron
---
tests/ofproto-macros.at |5 -
tests/ovs-macros.at |7 +++
2 files changed, 11 insertions(+), 1 deletion(-)
diff --git a/tests/ofproto-macros.at
This series makes it possible to include system-traffic.at tests into "make
check-offloads" tests.
The last patch of the series explains which tests are still not passing and
might need some more work.
I'll try to work on the remaining failing test cases or find someone who can
work on them.
> Thanks for the suggestion. These tests look like a good addition. If you
> want to send them formally to the mailing list, I can Ack them.
>
> thanks,
> Kevin.
Hi Kevin,
Sure, patch available on the mailing list here:
Bleep bloop. Greetings Ales Musil, I am a robot and I have tried out your
patch.
Thanks for your contribution.
I encountered some error that I wasn't expecting. See the details below.
checkpatch:
ERROR: Improper whitespace around control block
#154 FILE: lib/odp-util.c:8865:
Improve the test by adding and varying the MTU of a vhost port to check if
relevant mempools are created/(re)used.
Signed-off-by: Sunil Pai G
---
tests/system-dpdk.at | 61 +---
1 file changed, 58 insertions(+), 3 deletions(-)
diff --git
On Fri, Sep 2, 2022 at 4:19 PM Eelco Chaudron wrote:
>
>
> On 9 Aug 2022, at 9:17, Ales Musil wrote:
>
> > When the packet was traveling through patch port boundary
> > OvS would check if any of the actions is reversible,
> > if not it would clone the packet. However, the check
> > was only at
When the packet was traveling through patch port boundary
OvS would check if any of the actions is reversible,
if not it would clone the packet. However, the check
was only at the first level of the second bridge.
That caused some issues when the packet had gone
through more actions, some of them
26 matches
Mail list logo
