| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-cyborg - 2018-04-11 | |||
| 14:36:42 | zhuli | still work in progress on the os-acc lib, I had intended to refer the os-brick spec but seems there is not one for that, so I'm still looking inside the openstack/os-brick source code try to find the interaction between cinder and nova, it cloud help to find the functionality cyborg needed, will try to post a spec these days | |
| 14:37:04 | zhipeng | zhuli many thanks | |
| 14:38:10 | zhipeng | okey next | |
| 14:38:20 | zhipeng | #topic Vancouver Summit Prep | |
| 14:38:38 | zhipeng | anyone plan to attend the summit ? | |
| 14:38:48 | Coco | Coco from lenovo beijing | |
| 14:39:48 | zhipeng | anyone else ? | |
| 14:40:06 | NokMikeR | sorry transiting space and time, what was the question? | |
| 14:40:08 | Coco | We have a session with Rushil before, but seems he will not go there. | |
| 14:40:27 | zhipeng | Coco yes he'll skip that I guess | |
| 14:40:38 | zhipeng | NokMikeR will you be at Vancouver Summit ? | |
| 14:40:54 | NokMikeR | No I wont be going | |
| 14:40:54 | Coco | zhenghao, is also to be there. | |
| 14:41:08 | zhipeng | Coco great :) | |
| 14:41:17 | zhipeng | Li_Liu you got the funding yet ? | |
| 14:42:24 | Li_Liu | I am not sure yet, need to check with my boss... | |
| 14:42:31 | Li_Liu | no promises... | |
| 14:42:48 | zhipeng | okey we have told Steven Yuan about the cyborg project | |
| 14:43:00 | zhipeng | so I think if that came through you could claim the funding there | |
| 14:43:06 | zhipeng | but anyways | |
| 14:43:16 | Li_Liu | ok | |
| 14:43:58 | zhipeng | Again I will probably not be there, so Li Liu might be the only core team member present at the summit | |
| 14:44:19 | zhipeng | He will be representing me doing the Project Update, and also Forum sessions | |
| 14:44:19 | Coco | I knew a girl from ZTE, she sail he would attend the summit too. | |
| 14:44:26 | Coco | she | |
| 14:44:30 | zhipeng | Yes Yumeng will be there | |
| 14:44:59 | zhipeng | Okey speaking of Forum | |
| 14:45:09 | zhipeng | the deadline of forum topic proposal is Apr 15th | |
| 14:45:25 | zhipeng | plz feel free to propose any topic you want to discuss regarding cyborg | |
| 14:45:38 | zhipeng | Sundar already submitted one on FPGA | |
| 14:47:33 | zhipeng | #link https://wiki.openstack.org/wiki/Forum/Vancouver2018 | |
| 14:47:34 | Duth | Is the abstract of Sundar's paper available for review? | |
| 14:47:51 | zhipeng | Duth do you mean his spec or the forum topic proposal ? | |
| 14:48:54 | zhipeng | #link http://forumtopics.openstack.org/cfp/details/90 | |
| 14:49:00 | zhipeng | for example Sundar's proposal | |
| 14:49:37 | zhipeng | remember forum proposal does not require the consensus from the group, so if you got any ideas want to share or discuss, just feel free to directly propose it | |
| 14:51:21 | zhipeng | okey moving on | |
| 14:51:50 | zhipeng | #storyboad migration | |
| 14:52:02 | zhipeng | #topic storyboad migration | |
| 14:52:11 | Coco | can we use remotable accelerators in cyborg. | |
| 14:52:16 | Coco | ? | |
| 14:52:18 | zhipeng | please be aware that we are no longer using launchpad | |
| 14:52:34 | zhipeng | Coco we want to be able to do that | |
| 14:52:44 | zhipeng | and the os-acc will be part of the solution to make that happen | |
| 14:53:50 | zhipeng | back to topic | |
| 14:53:57 | kosamara | Not using launchpad for anything anymore? | |
| 14:54:14 | zhipeng | kosamara yes that is correct | |
| 14:54:29 | zhipeng | from now on, all the bug report, blueprint and so forth | |
| 14:54:35 | zhipeng | plz submit it onto storyboard | |
| 14:55:08 | zhipeng | #link https://storyboard.openstack.org/#!/project_group/77 | |
| 14:56:23 | zhipeng | okey | |
| 14:56:27 | zhipeng | #topic AoB | |
| 14:56:41 | zhipeng | any other business we want to discuss ? | |
| 14:56:44 | kosamara | I checked the vGPU spec | |
| 14:56:53 | zhipeng | kosamara plz continue | |
| 14:57:11 | kosamara | Comparing it with https://etherpad.openstack.org/p/Cyborg-Nova-Multifunction | |
| 14:57:45 | kosamara | it is basically implementing the D2 case in nova. | |
| 14:58:01 | NokMikeR | aob: an update or documentation on the install of cyborg with devstack would be great to see. | |
| 14:58:07 | zhipeng | but i think they were not able to actually fully do the D2 case right ? | |
| 14:58:27 | zhipeng | NokMikeR yes that would be part of the doc subteam's priority :) | |
| 14:58:47 | kosamara | Do you mean something specific out of D2? | |
| 14:59:19 | zhipeng | kosamara yes like Alex Xu commented | |
| 14:59:25 | NokMikeR | or a script that magically installs it is equally welcome. | |
| 14:59:37 | zhipeng | NokMikeR duly noted :) | |
| 14:59:43 | NokMikeR | lol | |
| 15:00:08 | kosamara | Yes, in their proposal it's a static configuration. | |
| 15:00:23 | zhipeng | yep | |
| 15:00:41 | kosamara | Each compute node needs to specify a mapping between gpu/vgpu_profile in a config file. | |
| 15:01:26 | kosamara | But I understood this static case to be the difference between D2 and B2 (orchestration). | |
| 15:02:58 | zhipeng | yes basically | |
| 15:04:24 | zhipeng | kosamara do you think if it is possible to improve the static configuration ? | |
| 15:05:50 | zhipeng | i mean in cyborg | |
| 15:06:08 | kosamara | The difference that I see between their proposal and Sundar's is that they are based on a static config, while Sundar relies on Placement traits. | |
| 15:06:56 | zhipeng | Sundar's proposal will need something like preferred trait | |
| 15:07:06 | kosamara | That means that Cyborg could implement Sundar's proposal as an alternative to the nova proposal, supporting a slightly different use case. | |
| 15:07:08 | kosamara | Yes. | |
| 15:07:20 | edleafe | Preferred traits are a long way off in Placement | |
| 15:07:32 | zhipeng | edleafe exactly | |
| 15:07:49 | kosamara | But D1/2 doesn't need preferred traits necessarily | |
| 15:08:31 | zhipeng | kosamara any chance you could write down some of your thoughts on this matter ? | |
| 15:08:41 | zhipeng | I think this could also be good to PoC :P | |
| 15:09:18 | kosamara | The flavor could require a particular vgpu profile as a trait, and then, as Sundar's proposal goes, Cyborg removes the incompatible traits from the resource provider that was picked by nova's scheduler. | |
| 15:10:31 | kosamara | Well a PoC is a bigger discussion. I'll start actually playing with vGPUs next week. RHEL 7.5 introduces support in kvm. | |
| 15:10:53 | zhipeng | sounds great :) | |
| 15:11:07 | kosamara | Initially I want to try out the nova proposal, and after that I can experiment with something else. | |
| 15:11:46 | kosamara | Does that fit with the rocky schedule? | |
| 15:11:57 | zhipeng | kosamara I think so | |
| 15:12:50 | Coco | Kosamara, we are doing the same thing now... | |
| 15:12:54 | kosamara | OK. So for a PoC we can talk again in ~10 days. I'll write some of these on Sundar's etherpad. Did you mean to write them somewhere else too? | |
| 15:13:15 | zhipeng | kosamara actually would be great to have an independant spec | |
| 15:13:15 | kosamara | Coco you mean trying vGPU with RHEL 7.5 | |
| 15:13:18 | kosamara | ? | |
| 15:16:05 | kosamara | I don't have any full-scale alternative proposal, I just compared nova's proposal with Sundar's. What would you expect in an independent spec? | |
| 15:16:39 | zhipeng | for example on the D2 case , or vGPU, what we could offer differently | |
| 15:16:48 | zhipeng | it does not need to be a full scale one | |
| 15:17:01 | zhipeng | since we will rely on Jianghua's spec as well | |
| 15:18:37 | kosamara | I think that Cyborg doesn't need to do 2 things for D2. The alternative to Sundar's is the nova spec (jianghua), which doesn't need cyborg. Cyborg adds the interplay with the extra Placement traits, which is nice for a little extra flex in D2. | |
| 15:20:07 | zhipeng | kosamara agree, what I meant is the overall vGPU data model could still follow Jianghua's, cyborg's own extra twist might come from your spec :) | |
| 15:20:45 | kosamara | ok... | |
| 15:20:59 | zhipeng | anyways, if you got the bandwidth :) | |