MaGate an interoperable, decentralized high-level scheduler Ongoing research at HES-SO/EIA-FR IANOS Meeting - Lausanne, July Ye HUANG, Pierre KUONEN Grid Group, Dept of Information and Communication Technologies, EIA-FR, Switzerland Pervasive Artificial Intelligence Group, Dept of Informatics, University of Fribourg, Switzerland
2 What’s MaGate MaGate [1] stands for “Magnetic Gateway”. GRID Resource management system MaGate is the grid level scheduler focusing on: Interoperability Decentralization Part of the Swiss project: SmartGRID framework [2]. Collaboration between EIA-FR and Uni. Fribourg Funded by the Hasler foundation
3 SmartGRID framework A decentralized grid scheduling framework. Supporting intelligent and interoperable grid resource management, upon the infrastructure information provided by swarm intelligence algorithms (ant technology).
4 SmartGRID framework: A Layered Architecture Loosely coupled layered architecture. Two layers and one internal interface. Smart Resource Management Layer Smart Signaling Layer Data Warehouse Interface
5 Smart Signaling Layer (SSL) Purpose: resource discovery network monitoring Approach: Ant algorithm Orchestration: Ant Nests community Two milestones have been achieved by UNIfr. Solenopsis middleware [3] Blåtant algorithm [4] for p2p path diameter shrinking (communication optimization).
6 Smart Resource Management Layer (SRML) Purpose: Grid level scheduling Interoperation Interaction with users and applications Access to external services Orchestration: MaGate schedulers community.
7 Data Warehouse Interface (DWI) To mediate the upper layer (SRML) and low layer (SSL). Loosely coupled information exchange mechanism. Both layers could benefit from DWI, by ignoring the implementation details of the each other.
8 SmartGRID Node (SG-Node)
9 SG-Node SG-Node stands for SmartGRID node. One GRID node portal GRID node Machine or group of machines providing computing resources through a unique access point Each SG-Node is constituted of one entity from each layer. One Ant Nest from SSL One Interface DW from SmartGRID internal interface. One MaGate from SRML
10 SG-Nodes community SG-Nodes collaborate to establish a dynamic and autonomic “SG-Nodes Community”. Behaving diverse roles depending on their positions in the infrastructure. Executor. Routers. Interface. Full functional.
11 MaGate Benefit from SmartGRID Decentralized grid scheduling mechanism. No central meta-scheduler. MaGate schedulers, instead of a single meta-scheduler, collaborate to provide grid scheduling. The grid infrastructure information discovery and monitoring. Thanks to ant technology - external component. Could be replaced by another IS approach Results promised to be robust, reliable and reactive. The possibility to contact the unknown grid scope. MaGates obtain grid information depending on ant system. Ants autonomously spread the network to discover unknown resource (nature ants’ food).
12 MaGate functionalities 1. Self-management 2. Grid community Management 3. Local resource management interface 4. External Grid services/components interface
13 Implementation approach Job allocation to local scheduler. DRMAA, SAGA, IANOS ?,… JSDL, BES Scheduling strategies. FIFS, Backfill. External scheduling policies. Cost model ?,… DWI Grasping the information Ant-Nest based existing Solenopsis middleware Others ?,… Information schema GLUE, …
14 Implementation approach (continued) Interoperability implementation. Core value and main focus of MaGate. Concerning the “community component”. Issues: Interoperation architecture. ’Teikoku’ [5] and ’Scheduling Instance’ [6] OGF GSA-RG: E.g. Simplest scheduling interop. use case Appropriate protocol/language. WS-Agreement series. Scheduling Description Language (SDL)
15 SmartGRID status Collaborative work: Ecole d'ingénieurs et d'architectes de Fribourg (EIA-FR). (Smart Resource Management Layer) University of Fribourg. (Smart Signaling Layer) A three-year project funded by Swiss Hasler Foundation, project Nr Approaching end of the first year for Smart Resource Management Layer.
16 Interest and Contribution Interested issues: As aforementioned, many of our interests are concerned with the workgroups that are also involved in IANOS. Our potential interests to IANOS: Joint standardization work (e.g. OGF) and experience sharing. Domain-specific use case/scenarios. Implementation of MaGate prototype using IANOS.
17 MaGate Using IANOS
18 Bibliography [1] Huang Y., Brocco A., Kuonen P., Courant M., Hirsbrunner B., SmartGRID: A fully decentralized Grid Scheduling Framework supported by Swarm Intelligence. Technical report, Working Paper 2008–06, Department of Informatics, EIA-FR, Switzerland, June 2008 [2] Hirsbrunner, B., Courant, M., Brocco, A., Kuonen, P.: SmartGRID: Swarm Agent-Based Dynamic Scheduling for Robust, Reliable, and Reactive Grid Computing. Technical report, Working Paper 2006–13, Department of Informatics, University of Fribourg, Switzerland, October 2006 [3] Brocco, A., Hirsbrunner, B., Courant, M.: Solenopsis: A framework for the development of ant algorithms. In: Swarm Intelligence Symposium, SIS, IEEE (April2007) 316–323 [4] Brocco, A., Frapolli, F., Hirsbrunner, B.: Shrinking the network: The bl°atant algo-rithm. Technical Report 08-04, Department of Informatics, University of Fribourg, Fribourg (April 2008) [5] Grimme, C., Lepping, J., Papaspyrou, A., Wieder, P., Yahyapour, R., Oleksiak, A., Supercomputing, P., Center, N., Waldrich, O., Ziegler, W.: Towards a standards-based grid scheduling architecture. Technical Report TR-0123, Institute on Resource Management and Scheduling, CoreGRID - Network of Excellence (December 2007) [6] Tonellotto, N., Yahyapour, R., Wieder, P.: A proposal for a generic grid scheduling architecture. Proceedings of the Integrated Research in Grid ComputingWorkshop (2005) 337–346
Thank you! Questions?