[ovs-dev] OVS on HyperV: meeting minutes for 8/20 IRC meeting

Nithin Raju nithin at vmware.com
Wed Aug 20 19:03:58 UTC 2014

Just sending out the meeting minutes for future reference.

Attendees: Alin Serdean, Samuel Ghinet, Ankur Sharma, Saurabh Shah, Nithin Raju

1. Review http://openvswitch.org/pipermail/dev/2014-August/044315.html
- VMware folks committed to reviewing this. Thanks cloudbase folks for working on this criticial patch. It was clarified that, VIF <-> VIF testing via VXLAN was tested, and we did see NBLs with multiple NBs each. VLAN based ping, which is much simpler, was not testing (and it is OK). This should be tested before checkin though.

2. Netlink attributes
- Discussion was around adding more validation code in the kernel netlink code. Everybody agreed that we'll add more validation code as we go.
- Cloudbase folks said they'll point out any validation code missing in the basic patches committed so far.

3. Status of netlink implementation: Status of userspace and kernel patches, Events and Packet receive infrastructure.
- #2 continued into #3.
- Ankur has posted a patch for the netlink parsing code. With this, the infrastructure pieces should be done (big ones at least).
- It was decided that once the infrastructure pieces are done, we'll fan out the work among multiple developers in VMware and Cloudbase.
- #Action-item: Nithin to create an issue for this, and fan out the commands to developers.
- Ankur's netlink parsing patch to be reviewed ASAP and enable the commands to be done.
- The advantages of ReadFile() v/s OVS_READ_IOCTL were discussed and it was concluded that, there's no immediate benefit of using ReadFile(). Same goes for WriteFile().
- Nithin clarified the recv_set(), recv_wait(), and recv() workflow and how it maps to overlapped events. Alin and Sam were satisfied with the explanation, and decided to go ahead with this model.
- #Action-Item: Nithin to post any details about recv_set(), recv_wait(), and recv() workflow that are missing in the design document.
- It was also clarified that events infrastructure is also similar to recv_set(), recv_wait(), and recv() workflow, and we'll have to map nl_subscribe_mcgroup() to this workflow.
- #Action-Item: Alin will be working further on the dpif-linux.c changes and posting them.
- #Action-Item: Alin to make changes to call DeviceControl(OVS_READ_IOCTL) instead of ReadFile(), and DeviceControl(OVS_WRITE_IOCTL) instead of WriteFile().
- The order of commands being focussed on are:
  * dp dump, and then vport dump, with the same workflow in userspace: dump start, dump, dump, dump, dump done.
  * Events
  * Packets

2. Persistent ports patch
- Nithin brought this up since there might be some overlap between this and the VPORT dump. But, we decided to table this for future since it was not clear as to what the overlap would be, and there was nothing concrete to investigate at this point. Sam pointed out the patches that have been posted.

3. P0 issues (if there's anything to discuss)
4. Other issues: when to start working on them
- Nothing to discuss here. All hands on deck to get the Netlink implementation done.

5.  OvsIpHelper vs the ARP method
- Sam pointed out that he wanted to discuss the advantages of OvsIPHelper v/s the ARP method. Discussions will be on the ML.

(Hope I didn't miss anything important).


More information about the dev mailing list