Hi Michal,
I've conducted some testing on v4 of the patch: specifically, wrt how OvS
reacts in the event of an error condition in the vhost construct/configure
functions.
For these tests I hardcoded the listed (DPDK) function to return an error
value, and observed how the following functions behaved, respectively:
- netdev_dpdk_vhost_construct (server mode only)
- netdev_dpdk_vhost_client_construct (client mode only)
- netdev_dpdk_vhost_client_reconfigure (client mode only)
Results are as follows:
#################################################################################################################################
| vhost function | vhost user mode | log
printed? | undesired behavior? | details |
#################################################################################################################################
| rte_vhost_driver_callback_register | server | yes
| no | n/a
|
| | client
| yes | no | n/a
|
+--------------------------------------+---------------------
+--------------- +-------------------------------------------------
+
| rte_vhost_driver_disable_features | server | yes
| no | n/a
|
| | client
| yes | no | n/a
|
+--------------------------------------+---------------------
+--------------- +-------------------------------------------------
+
| rte_vhost_driver_common_construct | server | yes
| no | n/a
|
| | client
| yes | no | n/a
|
+--------------------------------------+---------------------
+--------------- +-------------------------------------------------
+
| rte_vhost_driver_start | server
| yes | YES | vswitchd terminates
|
| | client
| yes | no | n/a
|
+--------------------------------------+---------------------
+--------------- +-------------------------------------------------
+
| vhost_common_construct | server
| yes | no | n/a
|
| | client
| yes | no | n/a
|
+--------------------------------------+---------------------
+--------------- +-------------------------------------------------
+
In summary, the only issue that I observed is when rte_vhost_driver_start
returns an error value to netdev_dpdk_vhost_construct, in which case, vswitchd
terminates - you might want to take a look at this.
Note that performance on the PVP path (with IP forwarding in the guest) was
marginally higher (~1.5%) with the 17.05 patch.
Finally, I also ran the usual sanity checks, and found no issues:
- checkpatch.py
- patch applies cleanly
- compiles with gcc
- compiles with clang
- sparse
- make check (no additional tests fail)
Thanks,
Mark
>From: [email protected] [mailto:[email protected]]
>On Behalf Of
>William Townsend
>Sent: Wednesday, June 7, 2017 3:57 PM
>To: Aaron Conole <[email protected]>
>Cc: [email protected]
>Subject: Re: [ovs-dev] [PATCH v4] Update relevant artifacts to add support for
>DPDK 17.05.
>
>Hi Aaron,
>
>On Wed, Jun 7, 2017 at 9:51 AM, Aaron Conole <[email protected]> wrote:
>
>> Hi Michal,
>>
>> mweglicx <[email protected]> writes:
>>
>> > Following changes are applied:>> > - netdev-dpdk: Changes required by DPDK
>> > API modifications.
>> > - doc: Because of DPDK API changes, backward compatibility
>> > with previous DPDK releases will be broken, thus all
>> > relevant documentation entries are updated.
>> > - .travis: DPDK version change from 16.11.1 to 17.05.
>> > - rhel/openvswitch-fedora.spec.in: DPDK version change
>> > from 16.11 to 17.05.
>> >
>> > v1->v2: Patch rework based on minor review comments.
>> > v2->v3: VHOST user client reconfiguration corrected.
>> > v3->v4: Patch is rebased against OVS master, minor
>> > rework based on review comments.
>> >
>> > Signed-off-by: Michal Weglicki <[email protected]>
>> > ---
>>
>> Since you got some comments from Mark, please describe the motivation
>> for this change with v5. Something explaining why to move off of the
>> LTS and onto this non-LTS version. I'm not opposed to moving - just
>> want to know what benefits it brings, and have that recorded as a log
>> in the commit history so that it can be referenced.
>>>> Thanks!
>>
>A big motivation would be to support new architectures like ARMv8 and
>Power9.
>I have had to move along with DPDK to get this to work.
>--Bill
>
>_______________________________________________
>> dev mailing list
>> [email protected]
>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>>
>_______________________________________________
>dev mailing list
>[email protected]
>https://mail.openvswitch.org/mailman/listinfo/ovs-dev
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
