http-3A__openvswitch.org_piperm
>ail_discuss_2015-2DOctober_019135.html&d=BQIGaQ&c=Sqcl0Ez6M0X8aeM67LKIiDJA
>XVeAw-YihVMNtXt-uEs&r=SmB5nZacmXNq0gKCC1s_Cw5yUNjxgD4v5kJqZ2uWLlE&m=OrWGS3
>tQ3PzCvpaif3pjUIuOxAY8nQCTY_8Q_Qs_Wj8&s=TeEaw1_lnbTm
Thanks for writing this up, it will definitely help many users.
I'm not sure the appropriate place for this is the OVS source tree
(given that's similar to README-native-tunneling.md), but this seems a
perfect candidate for a configuration cookbook
(http://openvswitch.org/support/config-cookbooks/
On 12/10/2015 18:02, "Stokes, Ian" wrote:
>Hi Daniele,
>
>Thanks for providing feedback, answers inline.
>
>> -Original Message-
>> From: Daniele Di Proietto [mailto:diproiet...@vmware.com]
>> Sent: Friday, October 09, 2015 6:53 PM
>&
Hi Ilya,
Sorry for the delay,
One question inline
On 11/09/2015 12:38, "Ilya Maximets" wrote:
>Currently tx_qid is equal to pmd->core_id. This leads to unexpected
>behavior if pmd-cpu-mask different from '/(0*)(1|3|7)?(f*)/',
>e.g. if core_ids are not sequential, or doesn't start from 0, or bo
Hi Ben,
I apologize for the delay, I can definitely review this.
Thanks,
Daniele
On 12/10/2015 19:09, "Ben Pfaff" wrote:
>Daniele, are you the right person to review this? If not, do you know
>who should review it?
___
dev mailing list
dev@openvsw
Hi,
Thanks for the patch, the implementation looks simpler than I expected.
General questions:
* It appears that we're using the srCTM, but
* We're only sending the green packets and dropping the rest.
* The `input_color` is always green.
Therefore
* `other-config:ebs` is ignored (isn't
Hi Joe,
I have a couple of minor comments inline
On 02/10/2015 22:16, "Joe Stringer" wrote:
>This patch adds a new action and fields to OVS that allow connection
>tracking to be performed. This support works in conjunction with the
>Linux kernel support merged into the Linux-4.3 development cyc
On 30/09/2015 17:59, "Joe Stringer" wrote:
>On 30 September 2015 at 09:03, Ben Pfaff wrote:
>> On Tue, Sep 29, 2015 at 01:40:34PM -0700, Joe Stringer wrote:
>>> When interacting with the local stack, the kernel may provide packets
>>> with existing ct state as they ingress OVS. These tests che
On 29/09/2015 18:56, "Ben Pfaff" wrote:
>On Tue, Sep 29, 2015 at 06:55:26PM +0100, Daniele Di Proietto wrote:
>> Newer kernels appear to require `bc` to build all the headers
>>
>> Also, alphabetize the package list
>>
>> Tested-at:
>>
On 29/09/2015 18:51, "Ben Pfaff" wrote:
>On Tue, Sep 29, 2015 at 10:47:56AM -0700, Joe Stringer wrote:
>> On 28 September 2015 at 20:27, Ben Pfaff wrote:
>> > On Mon, Sep 28, 2015 at 07:36:37PM +0100, Daniele Di Proietto wrote:
>> >> Recently some testc
Newer kernels appear to require `bc` to build all the headers
Also, alphabetize the package list
Tested-at: https://travis-ci.org/ddiproietto/ovs/builds/82757574
Signed-off-by: Daniele Di Proietto
---
.travis.yml | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/.travis.yml
On 29/09/2015 04:27, "Ben Pfaff" wrote:
>On Mon, Sep 28, 2015 at 07:36:37PM +0100, Daniele Di Proietto wrote:
>> Recently some testcases have been failing in travis because of a warning
>> related to the use of an L3 device (OpenVZ specific) inside the workers.
&
sudo access anymore, but we can
install packages with the apt plugin, and we shouldn't use root for
anything else
Tested-at: https://travis-ci.org/ddiproietto/ovs/builds/81764972
Signed-off-by: Daniele Di Proietto
CC: Joe Stringer
---
.travis.yml| 11 ++-
.travis/prepa
I tested the patch, but I wasn't able to reproduce your measurements.
On my test setup I noticed no difference in throughput for different packet
sizes/flow tables.
Could you describe your setup in more details?
I'd be happy to improve prefetching if it a simple change like the this.
Thanks,
D
Hi Joe,
thanks for sending this!
While doing some testing with my userspace connection tracker
on top of your series I encountered some small issues that
I was hoping you could squash in before pushing it to master.
None of the comments is supposed to be a blocker, we can address
them after merg
Thanks for taking the time to write this up, it is really detailed!
I think some of the information is more for developers than end users,
but I don't think this means it shouldn't be in INSTALL.DPDK.md.
I've fixed a couple of typos with a spell checker and applied this
to master.
Thanks
On 09/
On 10/09/2015 20:16, "Pravin Shelar" wrote:
>On Thu, Sep 10, 2015 at 10:02 AM, Daniele Di Proietto
> wrote:
>> DPDK mbufs contain a valid RSS hash only if PKT_RX_RSS_HASH is
>> set in 'ol_flags'. Otherwise the hash is garbage and doesn't
>> rel
Sorry for the delay.
There's still one problem with this patch:
when a non-DPDK port is added to the datapath, its txqs are not
added to the pmd threads.
Can you confirm the issue?
Thanks
On 10/09/2015 07:52, "Ilya Maximets" wrote:
>Ping.
>
>On 02.09.2015 14:44, Ilya Maximets wrote:
>> Curre
On 09/09/2015 18:56, "Pravin Shelar" wrote:
>On Wed, Sep 9, 2015 at 8:45 AM, Daniele Di Proietto
> wrote:
>> DPDK mbufs contain a valid RSS hash only if PKT_RX_RSS_HASH is
>> set in 'ol_flags'. Otherwise the hash is garbage and doesn't
>> rel
io Leitner
Acked-by: Kevin Traynor
Signed-off-by: Daniele Di Proietto
---
lib/dp-packet.h| 25 +
lib/dpif-netdev.c | 5 +++--
lib/netdev-bsd.c | 2 +-
lib/netdev-dpdk.c | 2 +-
lib/netdev-dummy.c | 2 +-
lib/netdev-linux.c | 2 +-
6 files changed, 32 insert
On 09/09/2015 17:56, "Ben Pfaff" wrote:
>On Wed, Sep 09, 2015 at 05:44:04PM +0100, Daniele Di Proietto wrote:
>> When building OVS with --enable-shared, -fPIC should be used in DPDK
>> CFLAGS. We used to add a custom option for this (CONFIG_RTE_BUILD_FPIC)
>> t
On 09/09/2015 17:38, "Ben Pfaff" wrote:
>We're starting to get build failures in travis starting from "dpdk: add
>support for v2.1.0", see
>https://urldefense.proofpoint.com/v2/url?u=https-3A__travis-2Dci.org_openv
>switch_ovs_builds_79313522-3A&d=BQIGaQ&c=Sqcl0Ez6M0X8aeM67LKIiDJAXVeAw-Yih
>VMN
of the file.
Furthermore, since vhost support is enabled by default in DPDK 2.1 and
vhost-user is OVS primary target, there's no need to customize the vhost
related option anymore.
Signed-off-by: Daniele Di Proietto
---
.travis/build.sh | 4 +---
1 file changed, 1 insertion(+), 3 dele
On 08/09/2015 18:12, "Traynor, Kevin" wrote:
>
>
>> -Original Message-
>
>> From: dev [mailto:dev-boun...@openvswitch.org] On Behalf Of Daniele Di
>
>> Proietto
>
>> Sent: Tuesday, June 16, 2015 7:39 PM
>
>> To: dev@openvswitch.
Leitner
Signed-off-by: Daniele Di Proietto
---
v1 -> v2:
* Added a comment above dp_packet_get_rss_hash()
* Added an OVS_UNUSED attribute on dp_packet_rss_valid()
---
lib/dp-packet.h | 13 +
lib/dpif-netdev.c | 2 +-
2 files changed, 14 insertions(+), 1 deletion(-)
diff --git
The patch looks good, I added you and Michal in AUTHORS and pushed
it to master.
Thanks
On 08/09/2015 07:56, "Puha, TimoX" wrote:
>Hi,
>
>> From: Puha, TimoX
>> Subject: [PATCH v3] dpdk: add support for v2.1.0
>>
>> Update relevant artifacts to add support for DPDK v2.1.0
>> - INSTALL.DPDK.md
On 28/08/2015 18:51, "Joe Stringer" wrote:
>On 28 August 2015 at 09:41, Daniele Di Proietto
>wrote:
>> I've tested it and it appears to correctly keep all the stats.
>>
>> Two comments inline, otherwise:
>>
>> Acked-by: Daniele Di Proietto
&
I've tested it and it appears to correctly keep all the stats.
Two comments inline, otherwise:
Acked-by: Daniele Di Proietto
Thanks for fixing this Alex!
On 28/08/2015 06:25, "Alex Wang" wrote:
>When dpdk configuration changes, all pmd threads are recreated
>and rx qu
With the older version of this series I was able to reproduce the problem
and this patch appears to fix it.
Acked-by: Daniele Di Proietto
Thanks!
On 28/08/2015 06:25, "Alex Wang" wrote:
>In the push_ukey_ops__(), when flow_del operation returns error, the
>'struct
The patch looks good, thanks.
Two comments below
On 26/08/2015 13:44, "Timo Puha" wrote:
>Update relevant artifacts to add support for DPDK v2.1.0
> - INSTALL.DPDK.md
> - acinclude.m4: Change DPDK library name
> - netdev-dpdk: Add 16 bytes extra padding to mbuf size to adapt to DPDK
>bug
> fi
Hi Alex,
sorry for the delay and thanks for taking care of this.
I couldn't experience any noticeable performance drop.
Acked-by: Daniele Di Proietto
On 22/08/2015 16:44, "Alex Wang" wrote:
>Sorry for the delay of pushing this and related dpdk patches,
>
>Want to s
The patch looks good to me.
I think that losing some packets when the thread are reconfigured
is unavoidable, considering that often the NIC must be stopped
to change the number of rxqs. Does this make sense?
Related to this, I've had an offline discussion with Ethan
and we realized that ukeys a
Sorry about the long wait.
I think this is the right direction, I like the idea of having a thread
local map of the txqs. It will allow us to assign the txqs in a more
sophisticated way.
I see some unit tests failing with this patch applied. Could you try
running the testsuite with 'make check'
It works both for userspace and kernel datapath, thanks!
Acked-by: Daniele Di Proietto
On 12/08/2015 22:01, "Joe Stringer" wrote:
>This test is skipped if the 'ip' command cannot interpret the vxlan
>'dstport'
>option; this is used as a proxy for de
Acked-by: Daniele Di Proietto
On 12/08/2015 22:01, "Joe Stringer" wrote:
>Depending on the kernel in use, manually removing a bridge during
>OVS_VSWITCHD_STOP can cause the kernel to send a route update that
>refers to the bridge device which is being deleted. OVS can
Acked-by: Daniele Di Proietto
On 12/08/2015 22:01, "Joe Stringer" wrote:
>Rather than saving all of the ping output to a file then checking at the
>end, check each ping and fail as soon as there is a connectivity
>failure.
>
>Signed-off-by: Joe Stringer
>---
>
Acked-by: Daniele Di Proietto
On 12/08/2015 22:01, "Joe Stringer" wrote:
>This allows arbitrary commands to be passed into the NS_EXEC macro to be
>executed within a namespace, including commands that have quotes and
>commands chained together.
>
>Signed-off-by: Jo
There has been some discussion lately about the status of the Open vSwitch
port to DPDK. While part of the code has been tested for quite some time,
I think we can agree that there are a few rough spots that prevent it from
being easily deployed and used.
I was hoping to get some feedback from th
. The ICMP related test in the conntrack branch has an
>instance of this.
>
>Maybe the right solution is to have two variants: NS_CHECK_EXEC() and
>NS_CHECK_EXEC_UNQUOTED(). I've seen this other places in OVS
>testsuite.
>
>On 10 August 2015 at 11:40, Daniele Di Proietto
>wrote:
>
My idea was to use ADD_BR inside the OVS_TRAFFIC_VSWITCHD_START()
parameters, but I'm fine with this approach too.
Acked-by: Daniele Di Proietto
On 08/08/2015 00:28, "Joe Stringer" wrote:
>This patch splits ADD_BR into two commands, so they can be used from
>differen
I guess you need this to use quotes("") inside 'command'.
One effect of this change is that if 'command' contains a pipe
(or &&, or ||) just the first command will be executed inside
the namespace. I'm not sure if it's a big problem. What do you
think?
On 08/08/2015 00:28, "Joe Stringer" wrote
I'm surprised as well that the tests were passing.
Sorry about this and thanks for finding out
Acked-by: Daniele Di Proietto
On 08/08/2015 00:28, "Joe Stringer" wrote:
>This was renamed. Surprisingly, the tests still pass without this,
>however the extra checks that
ation)
The userspace datapath is able to use system network interfaces via an
AF_PACKET socket.
Signed-off-by: Daniele Di Proietto
---
tests/.gitignore| 1 +
tests/automake.mk | 22 ++--
tests/system-common-macros.at | 3 ++-
tests/s
On 07/08/2015 19:15, "Joe Stringer" wrote:
>On 4 August 2015 at 11:00, Daniele Di Proietto
>wrote:
>> This series adds a new testsuite for the userspace datapath that runs
>> the already written kmod-sanity tests. The reason for this are
>> explained in
ecause they interact with system networking.
Suggested-by: Joe Stringer
Signed-off-by: Daniele Di Proietto
---
INSTALL.md | 7 +++
NEWS| 3 +++
Vagrantfile | 6 ++
3 files changed, 16 insertions(+)
diff --git a/INSTALL.md b/INSTALL.md
index 3dc16aa..9dac430 100644
--- a/INSTALL
Instead of repeating every time "ip netns exec ..." it is better to
introduce some macros.
Signed-off-by: Daniele Di Proietto
---
tests/system-common-macros.at | 27 +--
tests/system-traffic.at | 24
2 files changed, 33 inserti
t for running the system userspace
testsuite.
* Added ON_EXIT to remove port in ADD_VETH.
* Introduced NS_EXEC and NS_CHECK_EXEC to abstract the namespace
commands.
* Added mention in NEWS.
Daniele Di Proietto (4):
tests: Rename kmod-testsuite to system-kmod-testsuite.
tests: Introduce
The name makes more sense, especially with the addition of a userspace
system testsuite. No functional change in this commit.
Signed-off-by: Daniele Di Proietto
---
tests/.gitignore | 2 +-
tests/automake.mk | 24 -
tests/kmod-macros.at | 35
mespaces).
I can move some more linux specific code into the macros
(like the ip netns exec commands), but my goal for now would be to be
able to test the userspace datapath as well as the kernel module.
>
>On 4 August 2015 at 11:00, Daniele Di Proietto
>wrote:
>> @@ -32,10
On 05/08/2015 17:47, "Ilya Maximets" wrote:
>
>
>On 05.08.2015 19:26, Daniele Di Proietto wrote:
>>
>>
>> On 05/08/2015 16:42, "Ilya Maximets" wrote:
>>
>>> Sorry, I agree that example is incorrect. It is really no
ly one thread
> with core_id = 0. See dp_netdev_configure_pmd().
>
> So,
> pmd1 will call netdev_send(netdev=dpdk0, tx_qid= *pmd1->core_id* )
> pmd2 will call netdev_send(netdev=dpdk0, tx_qid= *pmd2->core_id* )
I agree, on current master they can't. I
On 05/08/2015 13:28, "Ilya Maximets" wrote:
>Currently tx_qid is equal to pmd->core_id. This leads to wrong
>behavior if pmd-cpu-mask different from '/(0*)(1|3|7)?(f*)/',
>e.g. if core_ids are not sequential, or doesn't start from 0, or both.
>
>Example(just one of possible wrong scenarios):
>
On 04/08/2015 02:38, "Joe Stringer" wrote:
>On 30 July 2015 at 11:21, Daniele Di Proietto
>wrote:
>> Acked-by: Daniele Di Proietto
>>
>> On 30/07/2015 00:52, "Joe Stringer" wrote:
>>
>>>The initial sanity test only checked IPv4 w
The file will be part of two different testsuites: one for the kernel
datapath and another for the userspace datapath. No functional change
in this commit.
Signed-off-by: Daniele Di Proietto
---
tests/automake.mk | 2 +-
tests/kmod-testsuite.at | 2 +-
tests/kmod-traffic.at | 107
space datapath is able to use system network interfaces via an
AF_PACKET socket.
Signed-off-by: Daniele Di Proietto
---
tests/.gitignore | 1 +
tests/automake.mk | 21 +++--
tests/kmod-macros.at | 19 ---
tests/traffic-common-macr
These macros will also be used by userspace datapath testing in
following commits. No functional change in this commit.
Signed-off-by: Daniele Di Proietto
---
tests/automake.mk | 1 +
tests/kmod-macros.at | 54 --
tests/kmod
testsuite is called `userspace-testsuite` and can be launched with
`make check-userspace`.
Daniele Di Proietto (3):
tests: Rename kmod-traffic.at to traffic.at.
kmod-macros: Move some code to traffic-common-macros.
tests: Add userspace-testsuite.
tests/.gitignore | 1 +
tests
On 31/07/2015 01:30, "Joe Stringer" wrote:
>On 30 July 2015 at 11:37, Daniele Di Proietto
>wrote:
>> I get a warning in the OVS log that causes this test to fail.
>>
>> It appears that when br0 is removed (in OVS_KMOD_VSWITCHD_STOP)
>> OVS gets a rtnet
| grep "transmitted" | sed
's/time.*ms$/time 0ms/'], [
3 packets transmitted, 3 received, 0% packet loss, time 0ms
])
+AT_CHECK([ip addr del dev br0 "10.1.1.100/24"])
+AT_CHECK([ip addr del dev br-ovs-p0 "172.31.1.100/24"])
+
OVS_KMOD_VSWITCHD_STOP
AT_CLEANUP
I&#
Acked-by: Daniele Di Proietto
On 30/07/2015 00:52, "Joe Stringer" wrote:
>We already queue the removal of the kernel module in OVS_VSWITCHD_START,
>via an ON_EXIT() call. This redundant removal also doesn't interact very
>well with usage of vports: If the datapath st
Acked-by: Daniele Di Proietto
On 30/07/2015 00:52, "Joe Stringer" wrote:
>The initial sanity test only checked IPv4 without IP fragments. This patch
>adds additional tests using IPv6 and VLANs with IP fragments and expands
>the existing test to be more strict.
>
>Si
This is interesting, thanks for the patch.
I'm definitely not a IVSHMEM expert, but I have a concern: what
happens if the secondary OVS process allocates or frees some
mbufs? (e.g because a packet is sent to multiple destinations or is
dropped)
I found this in the DPDK documentation:
http://dpdk.
Acked-by: Daniele Di Proietto
Thanks!
On 28/07/2015 07:55, "Ilya Maximets" wrote:
>Currently pmd threads select queues in pmd_load_queues() according to
>get_n_pmd_threads_on_numa(). This behavior leads to race between pmds,
>beacause dp_netdev_set_pmds_on_numa() starts
core 1 --> port 1, port 2
>core 2 --> port 2
>
>Fix this by starting pmd threads only after all of them have
>been configured.
>
>Cc: Daniele Di Proietto
>Cc: Dyasly Sergey
>Signed-off-by: Ilya Maximets
>---
> lib/dpif-netdev.c | 21 ++---
&
nfigured
>>>pmd
>>> + * threads on the same numa node. That's why we call
>>> + * 'dp_netdev_configure_pmd()' on all the threads and then we
>>>actually
>>> + * start them. */
>>> +for (i = 0; i <
That's a bad race condition, thanks for reporting it!
Regarding the fix, I agree that reloading the threads would
restore the correct mapping, but it would still allow
the threads to run with the incorrect mapping for a brief
interval.
How about postponing the actual threads creation until all
th
>Today I am working on confirming that this is the patch.
>
>Luis E. P.
>
>Sent from my iPhone
>
>On Jul 23, 2015, at 15:38, Ethan Jackson
>mailto:et...@nicira.com>> wrote:
>
>Ben, Justin, should this be backported? I'm not up on the policy at the
>moment.
there's no work to be done.
This commit fixes the problem by restoring the old values if
dpdk_eth_dev_init() fails.
Reported-by: Ian Stokes
Signed-off-by: Daniele Di Proietto
---
lib/netdev-dpdk.c | 9 +
1 file changed, 9 insertions(+)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-d
issue by retrying the device initialization
with a smaller number of queues, if a queue fails to setup.
Reported-by: Ian Stokes
Signed-off-by: Daniele Di Proietto
---
lib/netdev-dpdk.c | 100 +++---
1 file changed, 73 insertions(+), 27 deletions
The patch makes sense for branch-2.4 and master. Thanks!
Acked-by: Daniele Di Proietto
On 06/07/2015 22:41, "Mussar, Gary" wrote:
>Fix detection of vhost_cuse in dpdk rte_config.h
>
>Dpdk allows users to create a config that includes other config files and
>then
On 01/07/2015 11:49, "Timo Puha" wrote:
>New stats for vhost ports are rx_bytes, tx_bytes, multicast, rx_errors and
>rx_length_errors. New stats for PMD ports are rx_dropped,
>rx_length_errors,
>rx_crc_errors and rx_missed_errors. DPDK imissed packets are now
>classified
>as dropped instead of e
Hi Gary,
I wasn't able to apply the patch, it appears to be corrupt.
Would you mind resending it, perhaps using get send-email?
I also think that it would be useful to include the first
part of the email in the commit message, for future reference.
Otherwise the patch seems fine to me, thanks!
DPDK with vhost-user doesn't require libfuse, so we shouldn't link OVS
with libfuse unless DPDK is built with vhost-cuse support.
CC: Rapelly, Varun
Signed-off-by: Daniele Di Proietto
---
acinclude.m4 | 8 +---
1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/acin
On 24/06/2015 02:48, "Pravin Shelar" wrote:
>On Tue, Jun 23, 2015 at 11:42 AM, Ben Pfaff wrote:
>> Do you two have an opinion on this? If DPDK support is pretty solid now
>> then it makes sense to apply this to master and backport it to
>> branch-2.4.
>
>Personally I would like to have better
e:
>
>> >> On Mon, Jun 22, 2015 at 8:08 PM, Pravin Shelar
>>wrote:
>
>> >>> On Fri, Jun 19, 2015 at 11:24 AM, Daniele Di Proietto
>
>> >>> wrote:
>
>> >>>>
>
>> >>>>
>
>> &g
A vhost device, being a virtual interface, doesn't provide a valid RSS
hash. As a workaround we set the value to 0, which suggests the
datapath to recompute the hash in software.
Reported-by: Dongjun
CC: Traynor, Kevin
CC: Flavio Leitner
Signed-off-by: Daniele Di Proietto
---
lib/n
Signed-off-by: Daniele Di Proietto
---
lib/netdev-dpdk.c | 10 +-
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
index 3af1ee7..c03ff3d 100644
--- a/lib/netdev-dpdk.c
+++ b/lib/netdev-dpdk.c
@@ -93,8 +93,8 @@ BUILD_ASSERT_DECL
On 18/06/2015 23:57, "Traynor, Kevin" wrote:
>
>
>> -Original Message-
>
>> From: dev [mailto:dev-boun...@openvswitch.org] On Behalf Of Daniele Di
>
>> Proietto
>
>> Sent: Tuesday, June 16, 2015 7:39 PM
>
>> To: dev@openvswitch.
On 17/06/2015 19:08, "Flavio Leitner" wrote:
>On Tue, Jun 16, 2015 at 07:39:00PM +0100, Daniele Di Proietto wrote:
>> DPDK mbufs contain a valid RSS hash only if PKT_RX_RSS_HASH is
>> set in 'ol_flags'. Otherwise the hash is garbage and doesn't
>&
On 17/06/2015 14:18, "Traynor, Kevin" wrote:
>
>> -Original Message-
>> From: Traynor, Kevin
>> Sent: Wednesday, June 17, 2015 10:12 AM
>> To: Flavio Leitner; Daniele Di Proietto
>> Cc: dev@openvswitch.org; Gray, Mark D
>> Subject:
On 16/06/2015 18:27, "Flavio Leitner" wrote:
>On Mon, Jun 15, 2015 at 05:55:13PM +0000, Daniele Di Proietto wrote:
>> On 15/06/2015 12:16, "Traynor, Kevin" wrote:
>> >There is a dpdk patchset that contains a potential fix for this and
>>lots
>&g
t set the PKT_RX_RSS_HASH, forcing
OVS to compute an hash is software. This has a significant impact on
performance (-30% throughput in a single flow setup) which can be
mitigated in the CPU supports crc32c instructions.
Reported-by: Dongjun
Suggested-by: Flavio Leitner
Signed-off-by: Daniele
On 16/06/2015 07:40, "Pravin Shelar" wrote:
>On Mon, Jun 8, 2015 at 7:42 PM, Pravin Shelar wrote:
>> On Mon, Jun 8, 2015 at 6:13 PM, Wei li wrote:
>>> When tx queue is shared among CPUS,the pkts always be flush in
>>>'netdev_dpdk_eth_send'
>>> So it is unnecessarily for flushing in netdev_dpd
On 16/06/2015 09:55, "Gray, Mark D" wrote:
>>
>
>> Acked-by: Ethan Jackson
>
>>
>
>> One question I had for a future patch. Have you considered prefetching
>
>> more packets at once? I.E. 4 at a time or something?
>
>> That's how these things are typically written, though I don't know if
>>
On 16/06/2015 15:39, "Gurucharan Shetty" wrote:
>
>>
>> +global ps_type
>> +ps_type = ovs_vsctl("get Bridge %s datapath_type" % ps_name)
>> +if ps_type == "\"\"":
>> +ps_type = ""
>Else where in the code, this is handled by calling strip().
>e.g: port_type = ovs_vsctl("get In
With this commit, the VTEP emulator detects the datapath_type of the
bridge used as a "physical" switch, and creates subsequent bridges
with the same type. This allows ovs-vtep to work with the userspace
datapath.
Signed-off-by: Daniele Di Proietto
---
v2 - v3:
* Use strip('&q
With this commit, the VTEP emulator detects the datapath_type of the
bridge used as a "physical" switch, and creates subsequent bridges
with the same type. This allows ovs-vtep to work with the userspace
datapath.
Signed-off-by: Daniele Di Proietto
---
v1 - v2:
Applied Guru
n, Jun 15, 2015 at 11:28:12AM -0700, Gurucharan Shetty wrote:
>> > On Mon, Jun 15, 2015 at 11:09 AM, Daniele Di Proietto
>> > wrote:
>> > > With this commit, the VTEP emulator detects the datapath_type of the
>> > > bridge used as a "physical"
On 15/06/2015 19:28, "Gurucharan Shetty" wrote:
>On Mon, Jun 15, 2015 at 11:09 AM, Daniele Di Proietto
> wrote:
>> With this commit, the VTEP emulator detects the datapath_type of the
>> bridge used as a "physical" switch, and creates subsequent bridges
&
With this commit, the VTEP emulator detects the datapath_type of the
bridge used as a "physical" switch, and creates subsequent bridges
with the same type. This allows ovs-vtep to work with the userspace
datapath.
Signed-off-by: Daniele Di Proietto
---
vtep/ovs-vtep | 12 ++
ange has a much smaller effect (~0.5%
improvement)
Signed-off-by: Daniele Di Proietto
---
lib/dpif-netdev.c | 5 +
1 file changed, 5 insertions(+)
diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
index 5b82c8b..f13169c 100644
--- a/lib/dpif-netdev.c
+++ b/lib/dpif-netdev.c
@@ -3150,
On 15/06/2015 12:16, "Traynor, Kevin" wrote:
>
>> -Original Message-
>> From: Traynor, Kevin
>> Sent: Thursday, June 11, 2015 4:55 PM
>> To: Daniele Di Proietto; Flavio Leitner
>> Cc: dev@openvswitch.org; Gray, Mark D
>> Subject:
On 11/06/2015 15:23, "Flavio Leitner" wrote:
>On Thu, Jun 11, 2015 at 12:45:10PM +0000, Daniele Di Proietto wrote:
>>
>>
>> On 11/06/2015 04:25, "Flavio Leitner" wrote:
>>
>> >On Wed, Jun 10, 2015 at 03:13:21PM +, Daniele Di
On 11/06/2015 04:25, "Flavio Leitner" wrote:
>On Wed, Jun 10, 2015 at 03:13:21PM +0000, Daniele Di Proietto wrote:
>>
>>
>> On 10/06/2015 12:48, "Gray, Mark D" wrote:
>> >
>> >The vhost port won't generate an RSS hash becaus
On 10/06/2015 12:48, "Gray, Mark D" wrote:
>
>The vhost port won't generate an RSS hash because it is a virtual NIC.
>
>>
>
>> It doesn't cause a problem, just make the pkt fall into a slow path,
>>should we
>
>> fix it?
Thanks for investigating this. We should definitely fix it.
>
>> The fl
For other reasons I have (almost) the exact same commit in my local
repository!
Minor style nitpick below, otherwise:
Acked-by: Daniele Di Proietto
On 08/06/2015 17:36, "Ben Pfaff" wrote:
>Signed-off-by: Ben Pfaff
>---
> lib/flow.c | 18 +-
> 1 file
Thanks for figuring out this bug and for providing the patch!
I did a quick phy-phy test and the throughput looks the same.
I would be nice to fix the typo, as suggested by Mark. Otherwise
Acked-by: Daniele Di Proietto
On 05/06/2015 13:27, "Gray, Mark D" wrote:
>>
>
Thanks for the reviews!
On 03/06/2015 23:35, "Ethan Jackson" wrote:
>Acked-by: Ethan Jackson
>
>Merged.
>
>Ethan
>
>On Wed, Jun 3, 2015 at 10:45 AM, Flavio Leitner wrote:
>> On Wed, Jun 03, 2015 at 03:55:16PM +0100, Daniele Di Proietto wrote:
>>>
I sent a v3 that uses an alternative mapping computed when the
pmd thread is created. I hope it's clearer
Thanks!
On 02/06/2015 18:33, "Flavio Leitner" wrote:
>On Fri, May 29, 2015 at 12:42:00PM +0100, Mark D. Gray wrote:
>> From: Daniele Di Proietto
>>
>
Non pmd threads have a core_id == UINT32_MAX, while queue ids used by
netdevs range from 0 to the number of CPUs. Therefore core ids cannot
be used directly to select a queue.
This commit introduces a simple mapping to fix the problem: pmd threads
continue using queues 0 to N (where N is the numb
801 - 900 of 1317 matches
Mail list logo
