Chat transcript from room: tosca
2013-02-11 GMT-08:00
[08:04] Paul Lipton (Co-Chair) C1: Hi, running late. Will join soon.
[08:04] Matt Rutkowski (IBM): Interop SC meeting called to order at 10:04 AM US Central time
[08:07] Matt Rutkowski (IBM): Next weeks SC meeting will be cancelled
[08:08] Matt Rutkowski (IBM): expect Kavi notice
[08:08] Matt Rutkowski (IBM): Move to approve minutes from 2013-02-04, Paul 2nds, no objs.
[08:11] Matt Rutkowski (IBM): Derek: took us through his definition and thoughts behind it
[08:12] Matt Rutkowski (IBM): Derek: Load balancing and how they interact with our models is a key factor in thinking about why we need tier
[08:13] Matt Rutkowski (IBM): Derek: Load balancers laced in diff. avail. zones is a tech reason for needing tier concept
[08:13] Matt Rutkowski (IBM): Doug: do all nodes in a tier have to be same type?
[08:14] Matt Rutkowski (IBM): Derek: no, load balancers (actual or software) plus VMs of web servers would work together in a tier
[08:14] Matt Rutkowski (IBM): Doug: how do u dist. between a tier (grab bag) or a tier for a scaled out compute node
[08:14] Matt Rutkowski (IBM): Derek: tier is an abstraction of a compute node, plus other needed components
[08:15] Matt Rutkowski (IBM): Doug: may need to make that distinction more clear
[08:15] Matt Rutkowski (IBM): Derek: a tier is supposed to deal with a homogenous set of replaceable nodes (if one dies it can be replaced)
[08:15] Matt Rutkowski (IBM): Derek: tier is not a random bag of nodes
[08:16] Matt Rutkowski (IBM): Dale: is this a modeling decision? is a tier a node that may depend on a load balancer?
[08:17] Matt Rutkowski (IBM): Derek: when u impl. node balancing do you need to have to really have nodes that rep. those features, likely not
[08:18] Matt Rutkowski (IBM): Derek: a tier is an abstraction, that can be used to represent the logical parts without detailing all the wiring
[08:18] Matt Rutkowski (IBM): Paul: a tier is still a node?
[08:19] Matt Rutkowski (IBM): Derek: no it is a set of nodes
[08:19] Matt Rutkowski (IBM): Paul: how do u invoke a scale operation on a tier?
[08:19] Matt Rutkowski (IBM): Thomas: with the support for service template comp. we can have a tier as one node in a topology, but details in a sep. service template (abstraction)
[08:20] Matt Rutkowski (IBM): Thomas: the way we use tier in SugarCRM example, it does not quite fit this description
[08:20] Matt Rutkowski (IBM): Derek: we can be more explicit what it means
[08:21] Matt Rutkowski (IBM): Derek: if you have tier out of topology, the topologies end up not being equivalent (i.e. the logical results)
[08:21] Matt Rutkowski (IBM): Paul: how do u make the semantics work? Do u have operations at the tier level?
[08:21] Matt Rutkowski (IBM): Derek: aggregate ops u can do against nodes u can do against the tier as well
[08:22] Matt Rutkowski (IBM): Derek: tiers allow u to manage a set of nodes as a logical group
[08:23] Matt Rutkowski (IBM): Derek: semantics of a tier is groupwise, you change group behavior of all nodes within a tier at that level
[08:27] Matt Rutkowski (IBM): Richard: CMO consortium, in 2011, they made changes to spec. to include "group" as top-level meta language abstraction. That was a mistake, doing it as a node type in TOSCA, is a more elegant solution
[08:32] Matt Rutkowski (IBM): Matt: is hostedOn still accurate for tier?
[08:32] Matt Rutkowski (IBM): Derek: we currently use "hostedOn" for containment, we may need more refinement
[08:33] Matt Rutkowski (IBM): Thomas: in current SugarCRM model, does not fee like a natural fit for the use cases we have been discussing here today
[08:34] Matt Rutkowski (IBM): Thomas: service template based upon boundary defns. ; putting all these into a single topology does not seem correct. Perhaps we should
[08:35] Matt Rutkowski (IBM): Derek: assumed that to impl. complex tiers we would use boundary defns etc, but need to clarify these uses
[08:36] Matt Rutkowski (IBM): Thomas: SugarCRM, Tier is at bottom of topology, if u perform subst. it might appear in the middle of the topology so our usage given bound defns would be quite diff.
[08:41] Dale Moberg: Add a database tier as well as an app-ish tier?
[08:43] Matt Rutkowski (IBM): Matt: thanks to Dale for authoring guidance on use of IDrefs, Qnames, etc.
[08:53] Matt Rutkowski (IBM): Dale: took us through the document
[08:54] Matt Rutkowski (IBM): Matt: do we want to express consensus opinion of how to evaluate dependencies based upon rel. types
[08:54] Matt Rutkowski (IBM): Dale: would like to see if capturing these things is something we should do
[08:55] Matt Rutkowski (IBM): Dale: dependencies with cycles is an issue
[08:55] Matt Rutkowski (IBM): Derek: Need to think of real semantic validation (formal) how to interpret the data
[08:55] Matt Rutkowski (IBM): Dale: Happy to pursue the formal stuff
[08:56] Matt Rutkowski (IBM): Derek: for a declarative model there a concise set of assumptions
[08:56] Matt Rutkowski (IBM): Dale: ok with pursuing this; validation rules would be constraints on the model builder
[08:57] Matt Rutkowski (IBM): Derek: preserving names spaces has been an issue. non trivial models can be produced that others cannot consume
[08:58] Matt Rutkowski (IBM): Derek: interop testing these scenarios will bring these out. Neither are right or wrong, but need to have one way recommended on how we express certain things
[08:59] Matt Rutkowski (IBM): Derek: need to specify some constraints, are we going to have validation, profile based approach?

Matt asked Dale to provide a list of issues we need to address over time that tackle these processing concerns/rules/topics for initial discussion in two weeks time. Dale agreed.
[09:02] Matt Rutkowski (IBM): Matt moves to adjourn, Dale 2nds