Glue-schema 2002/01/09 ====================== Paul (me) Sergio Valentino Cristina Antonia Les Cottrel Brian Tierney Jenny Schopf Warren Matthews Eva Deelman Augusto Cuiffoletti Preamble ======== Quite a few new people. 1. current status ================= [Sergio] Sent a mail called "CE revision" which points to http://www.cnaf.infn.it/~sergio/datatag/glue/CE/revision.txt Changes to the schema for CEs. See CErevision-sergio.txt. 2. current issues/questions on the CE schema ============================================ 3. start discussion and brainstorming on how to define and model the GRID Network Element and how it can be used from the others Grid Services ========================================================================= [Augusto] Cristina sent out a doc by him: "For a definition of communication resources" He started thinking about why we need that network information. We have info about nodes - CE, SE, but no information about communication. From there we get: We could need information for every node to every other node. This is not good, because - there are large domains of well connected nodes - we need to make the measurements So, we look at it as domains, connected by links. A domain is a group of nodes, connected by good links. We assume that the nodes within the domain are connected to each other far better than to nodes outside the domains. Instead of end-to-end links we add specialised entities which make the network measuremtns and nothing else: these are called Theodolites. So, we have two new entities: the link between two domains and the Theodolites. This is the basic point of what "they" have been talking about in the last few months. They present it to GGF. A main point is that "internal" communication is far better than "external" communication. Not based on any single metric. All metrics are better inside than out. This is possibly quite difficult to guarantee. Define "connectivity" as a function of the metrics which reflects the quality of communication. [???] Not sure about the assumption that the connectivity inside the domain is better than that outside it. [Augusto] If you don't accept that assumption, then you lose all the organisation of the network, and then you cannot make these simplifications. [Warren?] Thinks it's very important because it is the only scalable solution, but also matches the way people organise networks. That is a spinoff benifit [Augusto] The main important point is that it reduces the amount of information required to be stored. [Cristina] Next step: Define what should be in the list of things measured and published for the domain-domain links. [Augusto] No, think we should decide what sort of model is acceptable. Once we have that, we can add any sorts of measurements. [M???] Thinks we have no choice but to use this model [Augusto] All the domains are at the same level - reduces complexity by a square, but not more. Domains could be associated with a site, or maybe not. e.g. two sites connected strongly could be the same domain. Not bound to administration or locality. [M???] Add an attribute to the CE and SE to indicate what domain they are a part of? [Augusto] Yep, this information is required somewhere. But where to put it? [Cristina] e.g. an RB finds a domain by its connectivity to another, then needs to find an SE in it. The mapping has to go somewher, clearly, but where? [Augusto] (gotten rid of NE, because it is overloaded). The mappings between Theodolite, Domain, SE, CE, Communication Entity (link between domains) should be static. Communication Entity c.f. Computing/Storage Entity Augusto's GGF presentation: sent to the list. [F???] Describe the GRISs. The Communication Element is what we are dealing with. Are the Theodolite GRISs? [Augusto] Well, the MDS structure can be different from the Domain structure. [Cristina] This can be added to the strawman. [Eva] We need to have a regular call. 21st Jan: finish compute: 6O'clock Italy Following week: start networking Don't mix CE and and Network