Minutes from the GSA RG meeting at GGF12. Sessions: - Sep 21, 9 a.m. - 10:30 a.m., Room: D002 o session was cancelled due to overlap with OGSA-EMS session - Sep 21, 3:30 p.m. - 5:00 a.m., Room: D002 Minutes taken by Uwe Schwiegelshohn (US), some additions by Ramin Yahyapour (RY) Session slides and minutes are available on GridForge: https://forge.gridforum.org/docman2/ViewCategory.php?group_id=133&category_id=783 ------------------------------------------------- Attendance list was distributed. About 45 people participated in the session. 0. Short Introduction and agenda (RY) - GSA-RG started at GGF 10 in March. - Goal is the identification of a Grid scheduling architecture. It is not the scope of this group to define scheduling algorithms. - It is expected that there will be not just a single scheduling strategy that suits the "Grid". Moreover, there will probably be many different scheduling approaches; many scheduling instances will interact with each other. - Therefore it is reasonable to identify common requirements of a Grid scheduling architecture, so that different implementations are possible. Grid Scheduling will have to interact with many services and rely on many protocols etc. However, currently many of the core services are not developed with Grid scheduling in mind. Therefore, a key task of this group is the identification what is necessary from a grid scheduler point of view. - Document roadmap: o identify use-cases (started GGF10, first version made available) o identify requirements on components and services for Grid scheduling (starting now, GG12) 1. Review of previous work on use cases (RY) - Some use-cases have been submitted and are available in the currently on GridForge available document. - Scope and setting of the use-cases differ significantly. - Reminder: The use-case document is not for describing existing Grid scheduling solutions from projects. Moreover, it is about specific application scenarios where Grid scheduling is used. The doc is necessary for understanding and concluding why something is necessary. ToDo: - the authors/editors might revise their contributions in regards to these remarks. - additional contributions are welcome. - the OGSA-EMS design team is also reviewing use-cases that can be helpful for GSA-RG 2. Philipp Wieder presented an example of a scheduling architecture based on one of the use cases. - see slides on GridForge 3. Identifying required services (RY) - Next milestones of the Group is: - Deriving requirements from the use cases - Identify required components/Services of a Grid scheduling architecture - At GGF11 the following first list has been compiled of grid scheduling-relevant components to which interaction and coordination is necessary: o Information Service o Job/Workflow Description o Requirement Description o Resource Discovery o Reservation o Monitoring/Notification o Job Execution o Security o Accounting/Billing o Data Management o Interface to Local RMS - Ramin went through a derived list with examples about expected functionalities from these components/services (see following comments). - Such a list must be extended and put in a document as a reference for discussion/exchange with other groups. 3.1 Information Service Relevant to Grid Scheduling: - coherent access to static and dynamic information - Dynamic information include data about planned or forecasted future events. o e.g.: existing reservations, scheduled tasks, future availabilities o not available by most current implementations - privacy concerns have to be considered. o not all of the information can be made available o but anonymized information would help also - information service should also cover resource types as network and data added during discussion: - access to historic information o useful for some schedulers strategies that make predictions 3.2 Job/Workflow/Requirement Description Relevant to Grid Scheduling: - access to information about the job specificy (what is the job?) - and job requirements (what is required for the job?) - need for workflow description o the Grid scheduler should be able to schedule a whole workflow o e.g. DAG formulation 3.3 Reservation Management/Agreement and Negotiation Relevant to Grid Scheduling: - Interaction between scheduling instances, between resource/agreement providers and agreement initiators (higher-level scheduler) - access to tentative information necessary - negotiations might take very long - individual scheduling objectives to be considered - probably market-oriented and economic scheduling needed such approaches often require many negotiation steps. - Need for combining agreements from different providers o coordinate complex resource requests or workflows o Maintain different negotiations at the same time o probable several levels of negotiations, agreement commitment and reservation 3.4 Accounting and Billing Relevant to Grid Scheduling: - access/interface to budget information o charging for allocation, reservations; preliminary allocation of budgets. - refunding in terms of resource/SLA failure, re-scheduling etc. 3.5 Job Management Services Relevant to Grid Scheduling: - Link to execution services to fulfill scheduled allocations - feedback on resource or job failures - interface to cancel or modify scheduled job executions - monitoring of agreement and execution status 3.6 Monitoring Services Relevant to Grid Scheduling: - Monitoring of resource conditions, agreements, schedules, program execution, SLA conformance, workflow 3.7 Interaction with Data Management Relevant to Grid Scheduling: - Access to information about the location of data sets - information about transfer costs - schedules about data transfers and data availability - coordination with network or other resources - Data management has many similarities with general grid scheduling: o access to similar services o similar tasks to execute o interaction necessary 3.8 Are services missing in this list? From the discussion in the session: - what about "provisioning"? o no clear definition o distinction between scheduling and job execution? - "Network Management"? o good point, added to the list. - "Historic Information"? o good point, addition to the requirements in information services - "Performance Prediction"? o good point, added to the list. 3.9 Next steps: - several volunteers for compiling the document about the required components/services - use the presented list as a starting point and extend it. - have a first draft ready until GGF13, March 05 4. Cooperation with OGSA-EMS design team - The OGSA-EMS team touches similar problems as GSA-RG. o identifying required services including Grid scheduling o however they are also splitting up the scheduling process itself; this is not part of GSA-RG as it is expected that there will be many different approaches on Grid scheduling. Therefore, it is not scope of GSA-RG to define actual scheduling algorithms/strategies. Moreover, it is anticipated to identify components and structure of a Grid scheduling architecture that allows many different implementations. Nevertheless, both groups can benefit from close cooperation to exchange experiences, use cases etc. Ramin tries to monitor OGSA-EMS discussion, but is not sure about the available time. Other volunteers would be welcome. 5. Conclusion and next activities: see above: - revise and extend use-case document - start document on required services and components