160 likes | 170 Views
Oracle for Physics Services and Support Levels physics-database.support@cern.ch Maria Girone, IT-ADC 6 April 2005. Outline. Database Services for Physics at CERN Scalable service for the LHC Update on service in 2005 Deployment of physics applications Conclusions.
E N D
Oracle for Physics Services and Support Levels physics-database.support@cern.ch Maria Girone, IT-ADC 6 April 2005
Outline • Database Services for Physics at CERN • Scalable service for the LHC • Update on service in 2005 • Deployment of physics applications • Conclusions Physics Database Services and Support Levels
Database Services for Physics • Mandate • Coordination of the deployment of physics database applications • Administration of the physics databases in co-operation with the experiments or grid deployment teams • Consultancy for application design, development and tuning • Organized the Developers Database Workshop in January 2005 • Involvement in 3D project and LCG Service Challenges • Provide database services for LHC and non-LHC experiments • Grid File Catalogs (RLS replacement expected in few months) • ~50 dedicated Linux disk servers • Sun Cluster (PDB01) • public 2-node for applications related to book keeping, file transfer, physics production processing, on-line integration, detector construction and calibration Physics Database Services and Support Levels
Challenges in 2005 • Service requirements from experiments will increase for LHC start-up preparation • Requirements still uncertain in some areas • Need to plan for a scalable service • Current infrastructure based on Sun cluster is already not sufficient to achieve required performance and application isolation • Linux based solution with ORACLEReal Application Cluster looks promising • Need to provide guaranteed resources to key applications • Need to identify them with experiments • Deploy them only after a validation phase Physics Database Services and Support Levels
Towards a scalable service for LHC • Looking into Oracle 10g RAC/Linux • Isolation – 10g ‘services’ and / or physical separation • Scalability - in both database processing power and storage • Reliability – automatic failover in case of problems • Manageability – significantly easier to administer than now • Coordinating a common work-plan across several IT groups • Hardware now in place and acceptance tested • RAC configuration and functionality tests going on now • Working on automated DB Server install integrated with s/w installation tools used for OS • Setting-up several RAC systems (so far, max 8-node RAC) • Target date for pre-production is summer 2005 • Implemented a stop-gap solution until then Physics Database Services and Support Levels
RAC Work plan • Main items view • RAC assembly & config tests: Q1 2005 • RAC optimization & stability test: Q2 2005 • Ramping up to production phase: Q3 2005 • Migration of all applications: Q4 2005 • Web site: http://cern.ch/it-adc/phydb/rac/ Physics Database Services and Support Levels
Stop-gap Service in 2005 • Several applications which either are high priority or large resource consumers have been identified • Performance monitoring in place to detect resource consumptions and changes in applications access patterns • Number of sessions, Server CPU used, number of server I/O operations • Resource priorities set by the experiments • Within the COCOTIME allocation • Allocated dedicated resources to service key applications • Provided so far one box/experiment on Oracle 10g • ALICE, ATLAS and LHCb already deployed, CMS under test • Once ready move to a defined slice of the new service cluster Physics Database Services and Support Levels
Planning for Application Deployment • Introduce a better defined deployment process to insure • Proper planning of database capacity (volume & CPU) • Insure the optimisation of key applications before production starts • Separate between two database application types • Resource consuming applications • Provide guaranteed resources • At first look: 10% of CPU, 10k sessions/day, 10% of max concurrent sessions, 10% max physical I/O • Standard applications • Smaller database applications which can run in a shared service Physics Database Services and Support Levels
Service Levels Layers • The new service will be based on Oracle 10g • Layered service implemented • Development Service • Code development, no large data volumes, no backup • Integration and Validation Service (for key apps) • Sufficient resources for larger tests and optimisation • Allocated together with consultancy manpower in advance: need 2 month notice • Production Service • full production quality service, including backup, monitoring services, on call intervention procedures • Monitoring to detect new resource consuming applications or changes in access patterns • The PDB Sun cluster will stay on 9i • Stop-gap resources and new cluster will be 10g Physics Database Services and Support Levels
A Typical Application Deployment Cycle • Development and definition of application requirements • Application development against the database development service • Once stable the application code and schema move to validation • Schema validation and optimization phase • Performance requirements are needed for key apps • How many operations? Which query mix? From how many concurrent clients? • Experiments should provide test work load • Work load is used to optimise application code and schema on a dedicated validation box • Additional diagnostics for dba and developers available • Work load & performance requirements are basis for resource planning • Resource Allocation & Planning • Optimised application with resource requirements is used to estimate h/w resources required for the service • Total database volume requirements are handled via COCOTIME Physics Database Services and Support Levels
Current state • 10g development and validation services and production dedicated resources in place • Now collecting requests from the experiments on the validation service • ATLAS prioritized list ready • CMS online applications list being prepared • Naming conventions being adopted (with input from Fermilab) • Proposal discussed in 3D project and available athttp://it-div-adc.web.cern.ch/it-div-adc/phydb/documents/Naming_Convention.pdf • We propose to hold another Developers Database Workshop focused on deployment in Q3 2005 Physics Database Services and Support Levels
Naming Conventions (1) • To simplify deployment of many applications on different service layers, we propose to adopt naming conventions • Project name of 18 characters max prefixed by experiment_ (VO_)Ex: LHCB_BOOKKEEPING, ATLAS_PIXEL_DT • Usernames should start with project name and the suffix • For integrationand production • owners: • none for production account with tables/views/etc • _INT for integration account with tables/views/etc • roles: • _W for account with write access to necessary tables/views/etc • _R for account only with read access • Others if requested for a fine grain control Ex:ALICE_TRACKER_INT_W ALICE_TRACKER_R Physics Database Services and Support Levels
Naming Conventions (2) • Tablespaces • Forintegrationandproduction: • _DATA for tables/etc • _INDX for indexes • Others if required for BLOBS or other data types Ex: ALICE_PIXELDT_DATA, ALICE_PIXELDT_INDX • Connection string • Today: one service name/project and a unique connection string in tnsnames.ora ATLASDB1 = (DESCRIPTION = (ADDRESS = (PROTOCOL = TCP)(HOST = ATLASRAC.cern.ch)(PORT = 1521)) (CONNECT_DATA = (SERVICE_NAME = ATLAS_SYS_PROD) ) • With 10g client, the possibility of avoiding tnsnames.ora files is being investigated (EZconnect user@host:port/service_nameEx: user@atlasrac.cern.ch:1521/ATLAS_SYS_PROD) Physics Database Services and Support Levels
Conclusions • Oracle deployment needs for physics are rapidly ramping up • Raising need for consultancy and deployment resources, to be handled in a coordinated manner • We propose to organize another Developers Database workshop focused on deployment in Q3 2005 • Validation service in place for optimization/tuning of prioritized resources • Experiments should maintain the list • Developing a new service on Oracle 10g to achieve scalability, availability and isolation • Stop-gap solution has been implemented to ensure guaranteed resources to production applications • This will be a critical year. Directing resources in key areas will be essential Physics Database Services and Support Levels
Service People and Responsibilities • Physics Database Services coordination: Maria Girone • Link people to experiments/projects: • Grid Deployment & Middleware: Miguel Anjo • ALICE: Jacek Wojcieszuk • ATLAS: Dirk Geppert / Ioannis Papadopoulos (POOL) • CMS & COMPASS: Giacomo Govi • LHCb & HARP: Andrea Valassi • Openlab Oracle Fellow: Marta Jakubowska-Sobczak • Validation service procedures: Ioannis Papadopoulos • Client side monitoring: Radovan Chytracek Physics Database Services and Support Levels