================================================ #opendaylight-group-policy: ODL-GBP-REQUIREMENTS ================================================ Meeting started by dconde at 20:01:00 UTC. The full logs are available at http://meetings.opendaylight.org/opendaylight-group-policy/2014/odl_gbp_requirements/opendaylight-group-policy-odl_gbp_requirements.2014-06-02-20.01.log.html . Meeting summary --------------- * https://meetings.webex.com/collabs/#/meetings/detail?uuid=M4NPPO6UULL1B0DFPK2HQ9VZX8-9VIB&rnd=563175.79385 (lenrow, 20:02:03) * goals (dconde, 20:07:50) * new invitation sent from Mike with WebEx (dconde, 20:10:06) * email came from dvorkinista (dconde, 20:10:19) * want scale and perf requirements (dconde, 20:10:50) * need three requirements that lenrow is showing (dconde, 20:12:12) * 1. how powerful the model is? 2. rate of change and 3. expose declarative intent NBI to expose traffic sterring and redirection (dconde, 20:13:34) * chrispriceab wants to avoid duality. (dconde, 20:15:13) * dvokiinista wants to avoid duplicating what is in affinity group and other projects (dconde, 20:16:24) * dvorkinista it can be rendered to affinity groups is needed (dconde, 20:17:07) * lenrow agrees. given he write original proposal for affinity project (dconde, 20:18:00) * dvorkinista shall we combine with affinity project? (dconde, 20:18:42) * lenrow says affinity does not have activity (dconde, 20:19:40) * chrisprice has concerns the if existing declarative API in ODL is not good enough if GBP there to fix it? (dconde, 20:23:35) * point 4. (dconde, 20:25:12) * dvorkinista wants transportable policy at a high level w/ operational policy that is enforced (dconde, 20:25:40) * dconde asked about alignment w/ Neutron (dconde, 20:27:36) * we want to create a general declarative model. (dconde, 20:27:59) * lenrow says putting ikeeping network state in black box is better (dconde, 20:28:52) * not on call but if you are talking about general declarative model, are you looking more at Congress? (regXboi, 20:28:57) * dvorkinista says neutron mixes some concepts. (dconde, 20:29:27) * alagalah says if we talking about steering, that's OK, but computation is different. (dconde, 20:31:58) * then it's about Congress (dconde, 20:32:05) * VMware says Congress is governance not orchestration (Heat) (dconde, 20:32:48) * says dvorkinista (dconde, 20:32:57) * questions to answer (dconde, 20:33:29) * seeing lenrow 's next slide (dconde, 20:33:43) * I presume so. It's lenrow who has it (dconde, 20:34:36) * who is the audience for GBP? (dconde, 20:35:07) * dvorkinista intent, governance and operational are three separate things. to be reasoned differently. (dconde, 20:35:34) * dvorkinista says policies were defined without identifiers. EP's enable items to be attached to it. (dconde, 20:37:35) * the group is persistent (dconde, 20:38:23) * dvorkinista policies are meant to be devoid of identifiers. (dconde, 20:38:42) * dvorkinista explained function of the end point registry. (dconde, 20:41:33) * lenrow then direct vs. indirect approach may not be relevant as shown in slide #5. (dconde, 20:42:16) * lenrow tallks about federation of controller. dvorkinista wants to solve it, but it's difficult. (dconde, 20:42:39) * dvorkinista says unlikely to be in first cut (dconde, 20:42:53) * other items to discuss (dconde, 20:43:21) * prove policies for regulatiotory compliance? asks lenrow re: financial customers. (dconde, 20:43:59) * dvorkinista that's tracability of rendering. (dconde, 20:44:42) * dvorkinista that's not enough. people from Univ of Illinois are looking at that. wish we can do that but that's beyond scope. (dconde, 20:45:19) * dvorkinista says it's a big deal, but worried "if you step wide, then your pants will run" (dconde, 20:46:10) * we can do some user-side valdation of use cases (dconde, 20:48:31) * dconde says we can have some canonical use casss we can show to customers. (dconde, 20:48:45) * sanjay making comments on analytics and provability. (dconde, 20:49:26) * it will items on reachability, routing, etc. that are beyond policy (dconde, 20:50:03) * dvorkinista once we get into L4-L7 it beomes very complex. (dconde, 20:50:35) * lenrow it would be great to get those validated from the "buy" side - as opposed to vendors (dconde, 20:51:13) * sanjay agrawal says that it could be done here (the analytics part) but not the probing part. (dconde, 20:51:45) * stuff like these EPs' will never talk to each other. That cab be analyzed. (dconde, 20:52:07) * dvorkinista says there can be many more use cases wehre it will be useful, beyond GBP. (dconde, 20:52:40) * audience slide (dconde, 20:54:41) * methodology (dconde, 20:54:51) * proposal 5 use cases (dconde, 20:54:58) * of 5 use cases, 3 chosen by ONF (dconde, 20:55:58) * multi-tier, NetVirt, DDOS traffic steerign are the three (dconde, 20:56:39) * next two are UC&C and IETF SFC (dconde, 20:56:52) * dconde asks itns NETVIRT a huge, broad topic? (dconde, 20:58:44) * lenrow says netvirt of connectivity is simple. when you layer on SLA, QOS, ACL then it's policy (dconde, 20:59:23) * dvorkinista and others saying we ought to drill down into what Netvirt means. (dconde, 21:01:05) * dconde says if we choose a more concrete subset, it's OK. (dconde, 21:01:22) * dvorkinista says lets put some bullet points under net virt (dconde, 21:01:52) * sanjay wants to add global enterprise app access as a use case (dconde, 21:02:53) * sanjay is taling about enteprise access control (dconde, 21:03:49) * users come based on location, net posture, etc. (regXboi, 21:04:05) * talking of more WAN use cases. lenrow says maybe chrisprice can add those? (dconde, 21:04:34) * sanjay says enterprise WAN, but Chris may want service provider use cases too. (dconde, 21:05:00) * chrisprice says let's look at enterprise first before doing service provider (100's of millions) (dconde, 21:06:25) * AGREED: that we ought to do enterprise first as priority (dconde, 21:06:36) * dconde asked about regXboi 's question on SFC (dconde, 21:08:08) * multi tier IAAS is similar to ONF (dconde, 21:08:34) * based on dvorkinista and Micorosft use cases. dvorkinista also talks of media companies' use cases tehre for IaaS (dconde, 21:08:57) * multi-tier has snapshot/coine (dconde, 21:09:20) * dconde asks about how we prioritize them? (dconde, 21:10:53) * lenrow says ONF had done some prioritization already to avoid too much analysis paralysis. (dconde, 21:11:16) * we ask whether these 5 are eough. Sanjay says these 5 covers the enterprise for the time being. (dconde, 21:12:57) * lenrow as long as we describe them well, we are OK, sine UC&C use case is complete. (dconde, 21:13:19) * we can spend a few hours over next two weeks to elaborate on the use-cases (dconde, 21:14:12) * sanjay asks should we start with the description, rather than PPT? (dconde, 21:14:32) * lenrow is open to methodology. (dconde, 21:15:07) * dvorkinista says let's not create bureaucracy for sake of it. (dconde, 21:15:29) * mickey_spiegel can we let people do doc ahead of time and meet to go over? (dconde, 21:15:59) * lenrow concerned we will add tons of use cases in call. (dconde, 21:16:20) * lenrow is willing to contribute some of thee use cases that have been examined for a long time. (dconde, 21:17:07) * dvorkinista expected format? (dconde, 21:17:26) * lenrow will do a sample based on ONF use-case doc. (dconde, 21:17:46) * lenrow will make sure ONF doc is public (vs. private ) (dconde, 21:18:17) * ACTION: lenrow will check if ONF use cases can be shared (dconde, 21:18:29) * uchau1 are we describing it in terms of precise GBP starting points, or? (dconde, 21:21:34) * dvorkinista let's descrbie the principles and then how it can be mapped into GBP. Quantitative /Qualittative requirements. (dconde, 21:22:01) * so use first order principlles first. says dvorkinista (dconde, 21:22:24) * dvorkinista next time let's do the UC&C (dconde, 21:25:58) * if UC&C is done, let's try others as examples. but uchau1 asks about where we look at mapping into GBP from model standpoint (dconde, 21:27:46) * uchau1 says diagram is enough, not JSON or YANG (dconde, 21:29:47) * AGREED: next time we step through UC&C and see if requierements are at correct level (dconde, 21:31:19) * not sure we can do that in two hours. (dconde, 21:31:42) Meeting ended at 21:33:22 UTC. Action items, by person ----------------------- * lenrow * lenrow will check if ONF use cases can be shared People present (lines said) --------------------------- * dconde (113) * regXboi (15) * lenrow (4) * odl_meetbot (4) * s3wong (2) * Bhushan_ (1) Generated by `MeetBot`_ 0.1.4