OGSA Teleconference - 21 April 2004 =================================== * Participants Ian Foster (ANL) Andrew Grimshaw (UVa) Bill Horn (IBM) Hiro Kishimoto (Fujitsu) Fred Maciel (Hitachi) Dejan Milojicic (HP) Andreas Savva (Fujitsu) Frank Siebenlist (ANL) Latha Srinivasan (HP) Jem Treadwell (HP) Apologies: Dave Snelling Minutes: Andreas Savva * Early discussion ** Teleconference minutes Apr.14 - no comments, approved ** Teleconference minutes Apr.19 - no comments, approved * May F2F update - Booked 1 big meeting room and 2 small ones (for sub-teams) - Might be better to just stay in one big room and split the space. Decide on the day. - Arranging a phone-bridge for large room. - Even though it is possible to call in remote participation is difficult so people are encouraged to attend in person. ** People attending who have not sent required registration ** details to ANL should do so ASAP - Expect a confirmation email if you have registered. - Make your reservation with the Guest House if have not done so yet (it is possible that only smoking rooms are available) - In preparation for the F2F prepare a folder/zip with all documents - Hiro has already created gridforge folder - Upload ALL reference and discussion material - Do NOT create subfolders. - Unfortunately gridforge folders are not real folders. The hierarchy cannot be downloaded in one go. So use descriptive names for files and put them all at the same level. (So that "Select All" and "Download zip" will get everything.) * GGF11 session plan update - Hiro has updated the GGF11 session plan, and created an online GGF11 folder Action: Add Ravi as co-owner of session 4 (PE) - Design team sessions to be scheduled in small rooms and somewhat inconvenient times, e.g., last day afternoon. - Security has two sessions (normal & design team). Action: Frank to check that this is ok for other security people. * OGSA glossary - "Grid Service" (Ravi and Jem will provide new definition) - Not available yet, skip. - "Failure" - replace "function" by "behaviour specification" - "failure detection" - as above - "failure recovery" - as above - "context" - Replace "event" (overloaded) with "conditions and circumstances when something takes place" - attaching credentials to "locus of execution" rather than "thread" (implementation) - Hosting Environment (HE) - Proposal to add other OSes, e.g., z/OS. Another example was .Net since J2EE is mentioned. - Proposal to delete product names from list, as a matter of policy. The problem is that doing so leaves very few if any examples. - No good idea so stay with product names for now (under protest). Action: Look for good alternatives to putting product names in HE definition. Action: Jem to revise glossary and re-post. Followup in next call. * Service level attainment loop (SLAL) discussion - There was an action for Jay, Andreas and Hiro to send out revised text. - Unfortunately there is no revised text. - Hiro sent out a related document a few hours before the call. The timing of sending out this document caused some confusion. - Hiro/Andreas clarified that this document is - a revised version of a portion of an earlier document (from a couple of calls back). - Does not have any input from Jay yet - It is aimed for the F2F, not this call. -It uses PE terminology to do a rendering of the SLAL. - Agreement that it is too early to discuss at this detailed level. (Not the intention anyway) - The figures look good otherwise - Andrew thinking of using them in the PE - Restatement that these interactions are not the only ones; it depends on the scenario illustrated. - Revisited key issue of last call: Is scheduling in provisioning and if so to what extent? (Using document in http://www-unix.gridforum.org/mail_archive/ogsa-wg/2004/04/msg00063.html) - Andrew's provisioning definition contains just the last 2 stages. What is the term applied to just those two steps? - Deployment & Configuration - According to this document they are part of a bigger picture. - Adding scheduling/allocating is one option to build this bigger picture. - Proposal to back off and try to decide what are the big components and then break them up into the smaller parts. Using discussion from last time these might be - local resource mgr - workload manager - provisioning - scheduler - Trying to draw a diagram of interaction without the use of a shared whiteboard is not doable. Action: Dejan to do one and share with other members. >Done. - Andrew talked about the scenario he is thinking (using PE terminology), going directly to resources. - If there is a ready resource configured correctly then just use it. - Otherwise transparently provide resources - JM drives the process - Dejan's perspective is that the scheduler gets the resources for the job - The PE perspective is that the scheduler (EPS) only goes as far as constructing the schedule. - Bill suggested doing a literature review since this topic is not that new. - Agreed that we are not using the same language and it looks unlikely to make progress just in calls. Need F2F time. - Andrew can see different ways of doing things. (Not committed to doing things in a specific way.) - Agreement that we have to clearly define and reach consensus on the objective first. - What functionality is encompassed; what problem is being solved. * Wrap up - Jem will check if HP can provide WebEx-like online whiteboard for OGSA call next week. - MS NetMeeting cannot work through firewalls so not everyone can use it - OGSA Telecon schedule is being updated after each call. https://forge.gridforum.org/projects/ogsa-wg/document/telecon_schedule/en - Next Monday call (Apr 26, 5-7pm CDT) is for Data and Security. - Next Wednesday call (Apr 28, 5-7pm CDT) is for Information Services and PE.