| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-cyborg - 2018-05-14 | |||
| 14:10:58 | dolpher | BTW, who else is attending this meeting? I see Sundar, shaohe, wangzhh | |
| 14:11:12 | Li_Liu | <--- me | |
| 14:11:17 | sundar_ | #link https://review.openstack.org/#/c/561849/ Cyborg agent/driver API spec. Please review | |
| 14:11:21 | dolpher | see ya :) | |
| 14:11:36 | sundar_ | I'll be there too | |
| 14:11:55 | wangzhh | I will attend with Jinghua Gao. | |
| 14:12:19 | sundar_ | rorojo: Rodrigo, you may want to say '#info rorojo' :) | |
| 14:12:39 | rorojo | #info rorojo | |
| 14:12:57 | shaohe_feng | welcome rorojo | |
| 14:13:06 | rorojo | Hi | |
| 14:13:16 | shaohe_feng | rorojo is an Intel FPGA experts. | |
| 14:13:23 | Li_Liu | welcome | |
| 14:13:23 | dolpher | @info dolpher | |
| 14:13:34 | dolpher | #info dolpher | |
| 14:13:56 | shaohe_feng | OK, let's start our topic. | |
| 14:13:58 | shaohe_feng | #topic current status of drivers | |
| 14:14:35 | shaohe_feng | Now cyborg support two drivers at present. Intel's FPGA, and SPDK | |
| 14:15:09 | shaohe_feng | So cyborg need to support more drivers. | |
| 14:15:36 | shaohe_feng | #action wangzhh will help to work on VGPU driver. | |
| 14:16:11 | Li_Liu | are there any xilinx folks here? | |
| 14:16:37 | shaohe_feng | Yes. any xilinx folks? | |
| 14:17:13 | shaohe_feng | they can help on xilinx FPGA driver. | |
| 14:17:29 | Li_Liu | I remember chuck and Dutch was attending our regular meetings | |
| 14:17:36 | Li_Liu | I will send email to them | |
| 14:17:46 | shaohe_feng | thanks Li_Liu | |
| 14:18:15 | shaohe_feng | #action Li_Liu will send an email to chuck and Dutch for xilinx FPGA drivers. | |
| 14:18:35 | shaohe_feng | for other drivers | |
| 14:18:59 | shaohe_feng | any one is working work smart nic, QAT etc? | |
| 14:20:19 | shaohe_feng | for intel FPGA driver, rorojo can help to improve it by OPEA API | |
| 14:20:21 | sundar_ | shaohe: We will be working on enabling a NIC | |
| 14:20:40 | rorojo | yes | |
| 14:20:52 | shaohe_feng | Great, thanks sundar_ | |
| 14:21:20 | sundar_ | shaohe: rorojo: It will be good to structure our work to go in paralle. Can the OPAE Pyhton API integration go in parallel to driver/agent API changes? | |
| 14:21:48 | sundar_ | I will help rorojo set up an environment and get started | |
| 14:21:56 | shaohe_feng | #action rorojo will improve intel FPGA driver. | |
| 14:22:23 | rorojo | Yes, I plan to ramp up on OpenStack this week | |
| 14:22:59 | shaohe_feng | great, any problem you an get help here. | |
| 14:23:12 | rorojo | thanks :) | |
| 14:23:41 | sundar_ | shaohe: Is anybody working on driver/agent API changes? | |
| 14:24:22 | shaohe_feng | sundar_: not at present. | |
| 14:24:43 | shaohe_feng | sundar_: we need a volunteer there. | |
| 14:24:48 | Li_Liu | what's the coresponding spec on that? | |
| 14:25:12 | sundar_ | #link https://review.openstack.org/#/c/561849/ | |
| 14:26:09 | shaohe_feng | sundar_: Li_Liu: maybe we need to list all specs, here. And let's others volunteer to take some task. | |
| 14:26:37 | sundar_ | #link https://review.openstack.org/#/q/status:open+project:openstack/cyborg All specs | |
| 14:27:52 | shaohe_feng | sundar_: you have 3 specs, will you code for all these specs？ | |
| 14:28:19 | rorojo | sundar: I believe then can go in parallel and they probably should | |
| 14:28:38 | shaohe_feng | rorojo: can you give a brief introduce for OPEA API? | |
| 14:28:39 | sundar_ | shaohe: The os-acc needs Nova changes. Just spoke to Nova folks for their review | |
| 14:28:52 | sundar_ | For agent/driver API, yes, I can do that | |
| 14:28:55 | shaohe_feng | rorojo: The status | |
| 14:29:19 | shaohe_feng | Is it ready for cyborg intel fpga driver? | |
| 14:29:24 | sundar_ | Nova/Cyborg scheduling: I can contribute, but I suspect multiple people will be involved | |
| 14:29:46 | rorojo | There is a pull request on github that describes the API | |
| 14:30:04 | Li_Liu | let me know if you need help. I prob will be focusing on the spec I wrote first | |
| 14:30:25 | rorojo | #link https://github.com/OPAE/opae-sdk/pull/413 | |
| 14:30:52 | shaohe_feng | rorojo: thanks. | |
| 14:31:02 | rorojo | yw | |
| 14:31:16 | sundar_ | shaohe and all: it may help to define some milestones, right? Shall we say that the first step is just discovery and assignment, which will help us get a basic VM up and running? | |
| 14:32:19 | shaohe_feng | sundar_: yes. agree. | |
| 14:33:08 | shaohe_feng | zhipengh[m]: ping. | |
| 14:33:33 | shaohe_feng | Li_Liu: do you know the code dead line? | |
| 14:33:57 | Li_Liu | better ask zhipengh.. | |
| 14:34:15 | sundar_ | shaohe: Good. So, there are three parallel tasks as I see it: driver/agent API changes, invoke update_provider_tree() to publish resources, and enhancements (like OPAE integration, anythine else?) | |
| 14:34:26 | Li_Liu | https://releases.openstack.org/rocky/schedule.html | |
| 14:34:58 | Li_Liu | Jun 04 - Jun 08 is when ms 2 ends | |
| 14:35:12 | Li_Liu | which is 3 weeks from now | |
| 14:35:26 | sundar_ | Yes, the specs need to be approved by then, IIUC | |
| 14:35:30 | Li_Liu | can we as least get the discovery working? | |
| 14:35:52 | Li_Liu | I thought the spec should be done by ms1 already | |
| 14:36:27 | shaohe_feng | Let do more reviews on spec first. | |
| 14:37:11 | sundar_ | Li_Liu: I was told that MS-2 is when specs freeze. That is true for other projects at least | |
| 14:37:13 | shaohe_feng | Jul 23 - Jul 27is Feature freeze. | |
| 14:37:21 | Li_Liu | I see | |
| 14:37:54 | Li_Liu | but feature freeze doesn't mean spec freeze tho | |
| 14:38:07 | Li_Liu | we should start implementation way before that | |
| 14:38:35 | Li_Liu | there's only 1 month between feature freeze and Rocky official release... | |
| 14:38:47 | sundar_ | Li_Liu: Agreed. If we can aim to get discovery done by MS-2, as you suggested, that would be good | |
| 14:39:31 | Li_Liu | since unlike scheduling, which involves nova folks, discovery should be totally self contained | |
| 14:39:50 | sundar_ | Yes | |
| 14:40:17 | sundar_ | Even scheduling can start a bit earlier: we can update the Nova provider tree without waiting for nRP support | |
| 14:40:32 | Li_Liu | yes, agree | |
| 14:40:41 | shaohe_feng | agree. | |
| 14:41:23 | shaohe_feng | Helloway: are you here? | |
| 14:42:06 | Helloway | yes] | |
| 14:42:31 | shaohe_feng | you have contributed a spdk driver. | |
| 14:42:42 | shaohe_feng | can you give us some introduce about it. | |
| 14:45:31 | shaohe_feng | ^ Helloway: | |
| 14:45:36 | Helloway | spdk driver can help cyborg to discovery the resources about spdk_based app(like nvmf_tgt, iscsi_target) | |
| 14:46:06 | sundar_ | helloway: Is there any specific hardware for this? | |
| 14:46:40 | shaohe_feng | Does it need a REST API? | |
| 14:47:03 | Helloway | include the nvme device and subsystems..and more | |
| 14:49:00 | sundar_ | helloway: So, is the expectation that Cyborg will discover the NVMe device as a PCIe device, and publish its properties as traits in Nova? | |
| 14:51:19 | Helloway | currently It does not have rest api? | |
| 14:51:32 | Helloway | currently It does not have rest api. | |
| 14:52:10 | shaohe_feng | will it report NVME device to placement? | |
| 14:53:39 | Helloway | sundar: not local pcie device, maybe it is more like nvme over fabric | |
| 14:53:40 | sundar_ | My understanding is, SPDK drivers are bit like DPDK poll mode drivers that run in user space. Something else needs to discover the PCI device. Helloway, please confirm if what you have is a user mode SPDK driver | |
| 14:55:08 | jiapei | I guess Helloway means something over fabric, through REST API | |
| 14:55:35 | jiapei | I'm also interested in this | |