14:03:13 <zhipeng> #startmeeting openstack-cyborg
14:03:14 <openstack> Meeting started Wed Sep  5 14:03:13 2018 UTC and is due to finish in 60 minutes.  The chair is zhipeng. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:03:15 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
14:03:18 <openstack> The meeting name has been set to 'openstack_cyborg'
14:03:25 <Li_Liu> #info Li_Liu
14:03:44 <wangzhh> #info wangzhh
14:03:46 <zhipeng> #info Howard
14:04:38 <wangzhh> Hi Howard, do u get the visa?
14:04:41 <wangzhh> :)
14:04:51 <dolpher> #info dolpher
14:05:22 <zhipeng> wangzhh yes got it lol
14:05:47 <zhipeng> #topic Stein PTG agenda crafting
14:05:55 <zhipeng> #link https://etherpad.openstack.org/p/cyborg-ptg-stein
14:06:07 <Sundar> #info Sundar
14:06:13 <wangzhh> Congratulations.
14:06:48 <zhipeng> as Li Liu mentioned, it seems that Nova team already planned a cross-project meeting with cyborg
14:06:53 <zhipeng> on Thursday right ?
14:07:13 <Li_Liu> yes
14:07:23 <Coco_gao> #info Coco_gao
14:07:39 <Li_Liu> Thursday "Cross Project Day" 11:10am - lunch: Cyborg/Nova cross project session
14:07:42 <Sundar> zhipeng: The main cross-project Nova/Cyborg needs to be on Cyborg time. Nova team will have a spill over time in case the discusions don;t complete
14:08:07 <Li_Liu> Is it in Cyborg's room or Nova's room?
14:08:19 <Sundar> We need to arrange for a session on Mon/Tue and let Nova team know ahead of time
14:08:52 <Sundar> Cyborg's room on Mon/Tue. The Thu slot is in Nova's room but that is a second follow-up
14:09:11 <Sundar> We do not expect all discussions to complete  in 1 hour session
14:09:37 <zhipeng> okey then what I would propose is that
14:09:42 <zhipeng> we concentrate on Monday
14:09:52 <zhipeng> and use Tuesday for spillovers
14:10:20 <Coco_gao> Agree
14:10:43 <zhipeng> I need to jump over the public cloud wg session as well, on Tuesday
14:10:53 <Sundar> zhipeng: Sure. We could arrange for the Nova session on mon afternoon? Hopefully that's enough time for folks to come in to the PTG venue
14:11:09 <zhipeng> That's good enough for me :)
14:11:18 <Sundar> Cool :)
14:11:32 <Li_Liu> So the main discussion will be happening on Mon and Tues?
14:11:39 <Sundar> Good to know you will be there :)
14:11:46 <zhipeng> The main one will be on Monday
14:11:50 <Li_Liu> ok
14:11:52 <Coco_gao> Do we have an outline about the meeting with nova?
14:12:04 <Li_Liu> I will take Monday off at least
14:12:26 <zhipeng> So some update, Li Liu will not make it to Denver unfortunately due to work conflict at company
14:12:27 <Sundar> I broadly wrote it as os-acc. The workflow itself is  still a matter of contention.
14:12:32 <zhipeng> and I got a late hail marry
14:12:37 <zhipeng> so I will stand in for Li
14:12:43 <zhipeng> to organize the meetings
14:12:48 <zhipeng> and dinners :)
14:12:54 <Coco_gao> Dinners~
14:13:22 <zhipeng> Ok let's go to the etherpad
14:13:30 <zhipeng> first organize the nova related topics
14:13:41 <Sundar> Li_Liu and all: please give your os-acc feedback in the spec before the PTG, in case you can't make it.
14:14:45 <Li_Liu> sure, will do.. I am kinda tied up recently. Will do some reviews in next couple days
14:15:09 <zhipeng> Sundar any idea for the Neutron cross discussion ?
14:15:13 <zhipeng> I really like that topic
14:15:28 <Sundar> zhipeng: Yes. Neutron has given a slot on Thu.
14:15:30 <zhipeng> shall we also invite Neutron guys over ?
14:15:43 <Li_Liu> can we propose it in Neutron's etherpad?
14:15:47 <zhipeng> Oh good, so we could talk them on Thu right ?
14:15:56 <Li_Liu> g8t
14:16:26 <Sundar> https://etherpad.openstack.org/p/neutron-stein-ptg Thu 3:15-4:15 pm
14:16:40 <Sundar> zhipeng: Yes
14:16:52 <zhipeng> nice !
14:17:26 <Sundar> There is also an Ironic/Neutron session on smart NICs. Not sure how far Cyborg will be involved there. But, since I am involved in that project, I'll be there too
14:17:29 <zhipeng> but we could align internally on Mon or Tuesday
14:17:41 <Sundar> Sure
14:18:18 <Sundar> As always, please give feedback in the etherpad early :)
14:19:39 <Sundar> We should have another session on broadening Cyborg scope to PCI devices. You might have seen Eric's question on whether we can handle IB for e.g. I think we can, without much architectural effort
14:19:59 <Li_Liu> Yes, that one
14:20:04 <Li_Liu> looking forward to it
14:20:12 <Sundar> The actual implementation, of course, can be left to whoever has the knowledge and resources for that :)
14:21:06 <Sundar> Also, how will Cyborg handle infrastructure offload? Like OVS offload?
14:21:27 <Sundar> Perhaps this is Neutron area, or may be Cyborg needs to be involved. One can argue both ways
14:21:45 <zhipeng> ya offload is another interesting area
14:21:57 <zhipeng> ovs/dpdk/ebpf offloading
14:22:23 <zhipeng> so what do we like for Monday morning ?
14:22:32 <zhipeng> retrospect of course
14:22:35 <zhipeng> and ?
14:23:09 <Sundar> Internal discussions -- lke db schemas?
14:23:38 <Coco_gao> after this, we can summarize what need to change in DB model and OVOs to make it more general.
14:23:47 <zhipeng> sounds good
14:24:14 <Sundar> RSD (composable arch) can also be Cyborg-internal to start with
14:24:26 <Li_Liu> Is Brian from Xilinx coming?
14:24:34 <zhipeng> i've no idea
14:25:02 <Li_Liu> I will follow up with him then, also the update on the Xilinx device drivers
14:25:10 <dolpher> Brian mentioned he could attend, but not confirmed with him
14:25:21 <wangzhh> We also need to disscuss how to handle different Arch. and system in cyborg.
14:26:22 <Sundar> wangzhh: can you elaborate? ^
14:28:01 <wangzhh> Such as x86, power(different) or libvirt, xen(different hypervisor).
14:28:47 <wangzhh> different Arch...  Sorry.
14:29:15 <wangzhh> Some operations like attach and detach
14:29:37 <Coco_gao> I suggest we discuss the "evolution" after "General Accelerator Management", It don't know whether this will have impact on DB.
14:30:03 <Sundar> wangzhh: Nova folks say that the instane attach and hypervisor-specific but not device-specific) ops should be left to Nova virt (and possibly os-acc but even that is debated)
14:30:04 <Coco_gao> what wangzhh said just now may also have impact on DB
14:30:46 <zhipeng> the general mgmt will be a long way to go
14:31:04 <zhipeng> and I want to make sure Eric Fried got time to join :P
14:31:04 <Sundar> Better to have Disaggregated servers as a separate topic
14:31:06 <shaohe_feng__> #info shaohe_feng_
14:31:34 <zhipeng> Sundar maybe under drivers ?
14:32:09 <Sundar> It will include representation etc.
14:32:34 <wangzhh> Sundar, It is vendor driver which handle these.
14:32:37 <Sundar> We can club it in any session :) as long as we know it will be wide-ranging
14:32:54 <zhipeng> ah alright :)
14:33:05 <Sundar> RSD is not vendor-specific :) Many OEM vendors support composable arch,
14:33:36 <Sundar> There is a large RSD+OpenStack effort outside of devices abd Cyborg
14:33:40 <Sundar> *and
14:34:13 <Sundar> We need to figure out how Cyborg fits in with Nova and others for that
14:34:38 <zhipeng> understood
14:35:04 <shaohe_feng__> Sundar: does ironic support to compose a RSD node, now?
14:35:33 <Coco_gao> @jiapei
14:36:41 <Sundar> The composition is separate from whether it is provisioned as a bare metal server. But I think Ironic + RSD is already on the cards
14:37:29 <Sundar> There is a RSD openstack client to compose nodes: https://github.com/openstack/python-rsdclient
14:39:31 <shaohe_feng__> yes, Do we need to integrate this client to ironic first?
14:39:58 <Sundar> Where do we fit infrastructure offload in these sessions?
14:40:50 <shaohe_feng__> Sundar: I have submit in another project called mogan to support RSD.  REF https://review.openstack.org/#/c/523052/
14:41:07 <Sundar> shaohe_feng: The client does not depend on Ironic, AFAIK. It works with RSD infra (Pod manager) to compose the node
14:41:55 <zhipeng> would be nice to migrate some of the mogan stuff over
14:43:25 <Sundar> shaohe: The RSD folks know about this effort, right?
14:43:26 <shaohe_feng__> yes. but we need a project to support the RSD server
14:44:23 <Li_Liu> can we leverage Cyborg to support it?
14:44:30 <shaohe_feng__> Sundar: I have not followed RSD in openstack for a long time.  With discuss with dolpher about the progress.
14:45:37 <zhipeng> okey moving on
14:45:48 <zhipeng> plz take a look at the agenda etherpad
14:46:01 <zhipeng> feel free to propose your topic
14:46:19 <zhipeng> next topic, team dinner :)
14:46:32 <zhipeng> Tuesday night works for everyone ?
14:46:36 <Li_Liu> Shall we assign time slot on the topics? tho I don't think people will follow anyways :)
14:46:59 <zhipeng> Li_Liu haha we could assign that
14:47:01 <Sundar> Tuesday night is a conflict with Intel dinner :)
14:47:10 <zhipeng> Monday night ?
14:47:37 <dolpher> I'll check if we can have the RSD folk to give a brief introduction on RSD+cyborg plan/design
14:48:38 <Sundar> Dolpher: I am the one dricing RSD+Cyborg within ntel :)
14:48:45 <Sundar> *driving
14:48:59 <Sundar> zhipeng: Mon night is fie by me
14:49:13 <Sundar> *fine
14:49:45 <zhipeng> everyone else ?
14:49:56 <zhipeng> Monday night for team dinner ?
14:50:19 <dolpher> Sundar: I mean ask Lin(working on RSD) to give a introduction, it's also fine if you volunteer to do that :)
14:50:23 <Li_Liu> damn.. you guys have fine :(
14:50:56 <dolpher> I'm fine with Mon night
14:51:05 <Coco_gao> It's better to have a introduction,~
14:51:16 <wangzhh> Zoom dinner with u. Li_Liu
14:51:28 <Sundar> Sorry to revert back to a previous point briefly: #link https://github.com/openstack/python-rsdclient/blob/master/doc/source/user/User_Guide.pdf The user composes a node first and then enrolls it in Ironic using the ironicclient. The two steps are independent
14:51:31 <Li_Liu> lol. that sounds fun
14:51:34 <Sundar> Lin is working with me :) Sure.
14:52:29 <dolpher> Sundar: I know that :)
14:53:41 <zhipeng> dolpher wangzhh Coco-cao
14:53:45 <zhipeng> team dinner
14:53:52 <zhipeng> is the time ok ?
14:53:56 <Coco_gao> OK
14:54:00 <dolpher> ok
14:54:18 <Sundar> ok
14:55:27 <zhipeng> Monday it is :)
14:55:30 <wangzhh> ok
14:55:54 <zhipeng> we could go to the Irish pub where the first cyborg team dinner happened :P
14:56:04 <zhipeng> one year ago, with Rushil and Justin
14:56:44 <Coco_gao> Sounds good.
14:57:20 <zhipeng> okey then
14:57:28 <zhipeng> #topic AoB
14:57:57 <zhipeng> any other business
14:58:43 <Coco_gao> nope
14:59:32 <zhipeng> okey see you guys in Denver :)
14:59:35 <Sundar> No. I'll try to get more specific questions for the Nova session before the PTG
14:59:48 <Sundar> Yes, see you there :)
14:59:49 <zhipeng> thx Sundar :)
15:00:08 <dolpher> looking forward to the team dinner then :)
15:00:21 <Sundar> BTW, a quick queston before we end
15:00:44 <Sundar> Did anybody hit an issue with devstack where Nova virt expects qemu > 2.5.0 on centos?
15:01:12 <Sundar> The default qemu pkg in centos is 2.0.0, so n-cpu does not start
15:01:45 <Sundar> It doesn;t look like centos provides qemu-system-x86_64 binary in a package with version >= 2.5.0
15:02:49 <dolpher> Have you switch to the cloud repo?
15:03:23 <Sundar> Do you mean qemu-kvm-ev or something else?
15:03:35 <Sundar> I am not using the qemu-*-ev repos yet
15:04:16 <dolpher> There  should be a repo for openstack, I can't remember the url
15:05:01 <Sundar> This one: yum install centos-release-openstack-queens ?
15:05:32 <shaohe_feng__> Sundar: xinran also have some RSD experience, also we have visit some RSD users in China.  You can aslo with xinran works on RSD
15:05:51 <Sundar> shaohe_feng: Good to know that. Thanks, sure
15:07:30 <zhipeng> #endmeeting