15:00:36 <ad_rien_> #startmeeting massively_distributed_clouds 15:00:37 <openstack> Meeting started Wed Jan 4 15:00:36 2017 UTC and is due to finish in 60 minutes. The chair is ad_rien_. Information about MeetBot at http://wiki.debian.org/MeetBot. 15:00:38 <openstack> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 15:00:40 <openstack> The meeting name has been set to 'massively_distributed_clouds' 15:00:43 <ad_rien_> #chair ad_rien_ 15:00:44 <openstack> Current chairs: ad_rien_ 15:00:47 <msimonin> o/ 15:00:53 <serverascode> hi 15:00:54 <ad_rien_> agenda 15:01:09 <ad_rien_> #link https://etherpad.openstack.org/p/massively_distributed_ircmeetings_2017 line 15 15:01:24 <ansmith> o/ 15:01:36 <ad_rien_> Hi folks 15:01:45 <ad_rien_> To start, I wish you all the best for 2017 ;) 15:02:02 <msimonin> thanks ad_rien_ :) 15:02:41 <denaitre_> o/ 15:02:47 <ad_rien_> #topic previous meeting action items 15:03:10 <HeleneCoullon> o/ 15:03:11 <ad_rien_> So regarding the last minute I cannot chair it unfortunately but I saw that you made few comments regarding the first slides we proposed 15:03:20 <ad_rien_> Hi HeleneCoullon denaitre_ 15:03:49 <ad_rien_> I updated the slides in order to take into account the comments you made 15:04:38 <ad_rien_> ansmith: I saw that you proposed to clarify the AMQP bricks in order to make a distinction between RabbitMQ/QPID/ZeroMq… 15:05:03 <ad_rien_> I just copied/pasted the slide and added a comment to ask you to update the figure according to the QPID bricks 15:05:09 <ansmith> ad_rein_: yes, the conversation was to have oslo.messaging be the service at the block level 15:05:27 <ansmith> and then show relationship to backend(s) at next level down 15:05:35 <ad_rien_> #link https://docs.google.com/presentation/d/1jJFZejZqgYDxu5FX4K8g3I5zQ87afnjYI4VSRSuCQ6U/edit#slide=id.p 15:05:49 <ad_rien_> so if you go on the slides 15:06:10 <ad_rien_> and you give a look at slides 2 and 3, you will see the comment. 15:06:26 <ad_rien_> May I ask you to update Slide 3 for QPID 15:06:28 <ad_rien_> ? 15:07:24 <ansmith> I will take the action to update the slide 15:07:54 <ad_rien_> ok great 15:08:08 <ansmith> ad_rien_: are you ok if I introduce oslo.messaging client/server blocks? 15:08:13 <ad_rien_> #action update slides for QPid and RabbitMQ 15:08:29 <ansmith> and then show relation to rabbit/amqp/zmq backends? 15:08:31 <ad_rien_> ansmith: yes please feel free to propose a new slide 15:08:44 <ansmith> ad_rien_: perfect 15:08:52 <ad_rien_> you folks from redhat are probably the best one to propose such a vision 15:09:10 <ad_rien_> ok 15:10:14 <ad_rien_1> or does anyone wants to add something on the current deployment scenarios we identified 15:10:32 <ad_rien_1> (sorry I'm facing network issues) 15:11:09 <ad_rien_1> so the next topic is to try to identify the pros/cons on the different scenarios 15:11:33 <ad_rien_1> Joe Huang added a comment in that direction this morning 15:11:42 <ad_rien_1> #topic Identify pros/cons of the different approaches 15:12:13 <ad_rien_1> So apart from the SPOF (single point of failure) of the current approach, do you see some issues ? 15:13:04 <ad_rien_1> From Inria, we are going to conduct performance experiments to identify network constraints in terms of latency/bandwidth 15:13:22 <ad_rien_1> for example in Scenario 1 (the simplest one) 15:13:51 <ad_rien_1> We would like to identify to what extent it is relevant from the network WAN constraint viewpoint. 15:14:32 <ad_rien_1> That is, we would like to see whether the network can be a strong constraint that can prevent such a deployment 15:14:46 <ad_rien_1> maybe latency issues, .... 15:15:02 <ad_rien_1> between nova-computes and services that are deployed on site 0 15:15:39 <msimonin> Like trying to find what are the minimum bandwidth and latency requirements to operate compute nodes far from core services ? 15:15:55 <ad_rien_1> Exactly msimonin 15:16:10 <ad_rien_1> It would be great to identify other experiments we can do to identify weaknesses or confirm advantages 15:16:50 <ad_rien_1> so folks any idea? ansmith? serverascode? 15:17:02 <serverascode> nothing from me, just lurking for now :) 15:17:07 <msimonin> :) 15:17:14 <ansmith> ad_rien_1: quick question on scenario 1, is nova-compute to services via L2 or L3? 15:17:51 <ansmith> e.g. any assumptions about openstack config 15:17:52 <ad_rien_1> L3 I would say 15:18:07 <ad_rien_1> since you go through the WAN 15:18:40 <msimonin> I guess we'll first experiment using the same L2 segment and try to figure out how to go through L3 15:19:11 <ad_rien_1> msimonin: can you elaborate please ? 15:20:10 <msimonin> From the experimentation we'll first go on the same L2 segment with network constraints and then got over L3 in the next iteration 15:20:20 <msimonin> just suggesting :) 15:20:24 <ad_rien_1> ok thanks 15:20:35 <msimonin> ansmith: is there any known deployment over L3 ? 15:20:35 <ad_rien_1> Ok I took some notes on our pad 15:21:22 <ansmith> ad_rien_l: some stretch site models extend the l2, with l3, there may be more configuration to consider of openstack services 15:21:48 <ansmith> ad_rien_l: I don't have direct experience there 15:22:05 <ad_rien_1> ansmith: I tried to write down this in the pad 15:22:13 <ad_rien_1> line 35 15:22:26 <msimonin> ansmith: what services do you have in mind ? 15:23:58 <ansmith> msimonin: thinking more generally that if there is an "internal mgmt" subnet, that the openstack api and messaging services likely reside on it 15:24:31 <ansmith> msimonin: so if WAN is l2, everything should just "work" 15:24:41 <msimonin> ansmith: sure :) 15:24:51 <ad_rien_1> so we have an action that is to identify lack of functionality according to the different scenario 15:25:12 <msimonin> I guess here some feedback/experience from telecom network would be beneficial 15:25:15 <ad_rien_1> How can we do that ? 15:25:40 <msimonin> Nobody from orange here ? : ) 15:26:02 <ad_rien_1> I mean we can at Inria take one or two scenarios and try to identify functional limitations (before performing performance experiments) but we need some helps for other scenarios 15:26:06 <ad_rien_1> That is the first possibility 15:26:31 <ad_rien_1> The second is to try to prioritize the different scenarios and start by focusing on the most important/representative/relevant one ? 15:26:57 <ad_rien_1> serverascode: AFAIK, we did not finalize our choice in the telecom WG? 15:27:00 <ad_rien_1> did you ? 15:27:25 <serverascode> in terms of distributed clouds? no don't think we finalized anything 15:27:35 <ad_rien_1> #action Identify functional needs/limitations for each scenario 15:27:52 <ad_rien_1> I mean in terms of what can be the minimal NFV deployment 15:27:58 <ad_rien_1> ? 15:28:14 <msimonin> yes 15:28:16 <ad_rien_1> By merging our effort we can probably progress faster 15:28:34 <serverascode> ah ok, no we haven't finalized anything, just agreed to start working on a document to define it 15:28:40 <ad_rien_1> ok 15:28:49 <serverascode> in terms of a "generic NFV platform based on OpenStack" 15:28:58 <ad_rien_1> yepp 15:29:39 <ad_rien_1> We met some folks form create-net during the cloudcom conference in december 15:29:59 <ad_rien_1> the guys investigate the scenario 1 (i.e. slide 0) in a context of edge computing 15:30:10 <ad_rien_1> (i.e. they have a large cloud and few remote compute nodes). 15:30:22 <ad_rien_1> For the moment, they only performed paper works 15:30:59 <ad_rien_1> and told us that they are interested by following our progress. I hope they will join us. but once again, if we can combine our effort, it would be valuable for all of us 15:31:22 <ad_rien_1> for Fog/Edge scenario 1 looks to be the simplest 15:31:31 <ad_rien_1> I do not know whether it makes sense for NFV operators 15:33:00 <ad_rien_1> ok so getting back to my initial question: how can we move forward. Can we sort the different scenarios? 15:33:07 <ad_rien_1> and focus on the most important one? 15:33:49 <ad_rien_1> #action prioritize the different scenarios 15:34:11 <msimonin> if some people are interested in scenario 1 we should start by this one 15:34:37 <ad_rien_1> +1 15:34:37 <msimonin> => compute far from core services 15:35:02 <ad_rien_1> this will also enable the evaluation of other messaging solutions 15:35:24 <ad_rien_1> so let's focus on this first scenario. Is it ok for everyone? 15:35:29 <ansmith> +1 15:36:39 <ad_rien_1> #info everyone agrees to start with scenario 1 15:36:41 <kgiusti> +1 15:36:45 <msimonin> :) 15:37:00 <ad_rien_1> so let's move to the next topic 15:37:07 <ad_rien_1> #topic ENOS status 15:37:15 <msimonin> yeah ! 15:37:26 <msimonin> so 15:37:45 <ad_rien_1> msimonin: can you please briefly introduce what is ENOS and how it can be helpful to investigate the different scenarios ? 15:37:53 <ad_rien_1> I give you 3 minutes :) 15:38:04 <msimonin> thanks :) 15:38:19 <msimonin> #link https://github.com/beyondtheclouds/enos 15:38:38 <msimonin> we've build a tool to evaluate openstack in various connfiguration and topology 15:38:49 <msimonin> there is a white paper somewhere 15:38:53 <msimonin> let me find the link 15:39:36 * msimonin searching :) 15:39:56 <msimonin> #link https://hal.inria.fr/hal-01415522v2 15:40:01 <msimonin> found it :) 15:40:17 <msimonin> so we already made some scalability tests 15:40:23 <msimonin> presented at Barcelona 15:40:41 <msimonin> and we plan to extend to support multiregion deployment as well as 15:40:46 <msimonin> network constraints enforcement 15:41:08 <msimonin> so basically we'll be able to evaluate the different scenarios 15:41:14 <msimonin> (hopefully :) ) 15:41:49 <msimonin> I think that's all for the overview ad_rien_1 :) 15:41:56 <ad_rien_1> thanks 15:42:21 <ad_rien_1> Just to complete, Scenario 1 can be evaluated, Scenario 2 is cells (for the moment we do not consider it as the cell v2 refactoring is under heavy development) and finally, Scenarios 3 can be evaluated. 15:42:25 <ad_rien_1> is it correct? 15:42:30 <msimonin> yes 15:42:37 <ad_rien_1> thanks 15:43:01 <ad_rien_1> so the next step regarding ENOS is to investigate how we can deploy other messaging solutions such as QPiD 15:43:30 <ad_rien_1> ansmith: kgiusti: are there ansible playbooks to deploy QPID ? 15:43:42 <ad_rien_1> (actually ENOS is leveraging kolla and ansible) 15:44:03 <ansmith> ad_rien_l, msimonin: there is puppet module but ansible shouldnt take too much 15:44:34 <msimonin> could you link the puppet module ? 15:45:14 <ad_rien_1> or even better, if you can provide some ansible playbooks if doing the port is not too time consuming ;) ? 15:45:44 <ansmith> puppet module reference in etherpad 15:45:52 <msimonin> ansmith: ack 15:46:32 <ansmith> ad_rien_l: will assess ansible effort 15:46:38 <ad_rien_1> thanks 15:47:02 <ad_rien_1> #action ansmith will give a look to ansible playbooks for QPID 15:47:23 <ad_rien_1> is there any question regarding ENOS? 15:48:06 <ad_rien_1> ok so let's move to the next topic 15:48:39 <ad_rien_1> #topic expected capabilities for Massively distributed/Fog/Edge clouds 15:48:57 <ad_rien_1> as I mentioned I was not chairing the last meeting 15:49:09 <msimonin> Last time we started to discuss migration with someone from orange 15:49:17 <ad_rien_1> so based on the logs it looks like this point was not sufficiently discussed 15:49:22 <ad_rien_1> ok 15:49:31 * ad_rien_1 is looking for the logs. 15:49:40 <msimonin> but he seems not here today 15:50:54 <ad_rien_1> #action do not forget to discuss about migration between remote sites with Orange folks 15:52:04 <ad_rien_1> Can eveyone access the pad: https://etherpad.openstack.org/p/massively_distributed_ircmeetings_2017 ? 15:52:07 <ad_rien_1> #link https://etherpad.openstack.org/p/massively_distributed_ircmeetings_2017 15:52:33 <ad_rien_1> I just copied/pasted few ideas we have at Inria regarding this topic. 15:52:50 <ad_rien_1> can you go through the notes and put +1 if you believe it makes sense. 15:54:35 <ad_rien_1> We have 5 minutes before ending the meeting? 15:54:44 <ad_rien_1> any comments on the expected capabilities? 15:55:22 <ad_rien_1> it seems not so I propose to switch to the last topic. 15:55:34 <ad_rien_1> and we will come back on this point later. 15:55:39 <ad_rien_1> #topic Open Discussion 15:55:57 <ad_rien_1> So from my side, just two remarks (already written in the pad) 15:56:15 <ad_rien_1> 1./ We talked about folks from RedHat that are investigating Fog/Edge computing challenges. 15:56:35 <ad_rien_1> ansmith: kgiusti any chance to talk with them during one of our meetings? 15:56:50 <ad_rien_1> I think it would be really great to exchange our point of views. 15:57:10 <ansmith> ad_rien_l: I reached out to them prior to break and they were very interested, I will follow up 15:57:39 <ad_rien_1> #action ansmith will check whether other folks from redhat can participate to the WG. 15:57:44 <ad_rien_1> thanks ansmith 15:57:52 <ad_rien_1> we can discuss my second point later 15:58:00 <ad_rien_1> we have just two more minutes. 15:58:09 <ad_rien_1> So I prefer to leave the floor to you guys 15:58:19 <ad_rien_1> anything from your side ? 15:58:34 <serverascode> nohting from me 15:58:55 <ad_rien_1> ok thanks for attending the meetinhg 15:59:04 <ad_rien_1> talk to you in two weeks. 15:59:13 <ad_rien_1> bye 15:59:22 <serverascode> thanks! 15:59:25 <ad_rien_1> #endmeeting 15:59:25 <ansmith> thanks 15:59:47 <ad_rien_> #endmeeting