Скачать презентацию David Foster LCG For PEB Meeting 9 4 02 Fabric Скачать презентацию David Foster LCG For PEB Meeting 9 4 02 Fabric

28544c086b430e2bd20fa472ec651ec5.ppt

  • Количество слайдов: 13

David Foster, LCG For PEB Meeting 9/4/02 Fabric Planning • Extension of scope of David Foster, LCG For PEB Meeting 9/4/02 Fabric Planning • Extension of scope of the Fabric Management area to “Fabrics” – Fabric Management – CPU Farms – Disk Farms – Mass Storage • Look at the fabric solution as a whole.

David Foster, LCG For PEB Meeting 9/4/02 High Level Requirements • Meet data/computing challenge David Foster, LCG For PEB Meeting 9/4/02 High Level Requirements • Meet data/computing challenge requirements (needs more definition) • Operate with minimum manpower (operations and engineers). Needs targets. • Repurpose nodes between different standard environments. • High level of robustness and availability (average node availability > 99% ? )

David Foster, LCG For PEB Meeting 9/4/02 General LCG Milestones • June 2002 – David Foster, LCG For PEB Meeting 9/4/02 General LCG Milestones • June 2002 – Establish “stable testbed” with EDG software • September 2002 - Scale to 200 machines • October 2002 – Evaluate success • November/December – Glue final deliverable • January/March LCG Software integration • March 2003 - LCG 1. 0 Production System

David Foster, LCG For PEB Meeting 9/4/02 Fabric Management (FM) High Level Milestones • David Foster, LCG For PEB Meeting 9/4/02 Fabric Management (FM) High Level Milestones • July 2002 – Simple Automation – EDG “stable testbed” • September 2002 – Semi-Automated – Scale to 200 nodes • March 2003 – Full Automation

David Foster, LCG For PEB Meeting 9/4/02 FM Milestones - Detail • Simple Automation David Foster, LCG For PEB Meeting 9/4/02 FM Milestones - Detail • Simple Automation – New version of LCFG with some enhancements. • Semi-Automation – Initial use of HLD tools together with existing clients. • Full automation – – – New clients implemented. Integrated with monitoring system. New Scalable updaterpms. Kickstart file generation from HLD Old solutions (Sue/Bis) fully deprecated.

David Foster, LCG For PEB Meeting 9/4/02 Open High Level FM Issues • Ability David Foster, LCG For PEB Meeting 9/4/02 Open High Level FM Issues • Ability to re-purpose nodes flexibly (change complete environment automatically). • Policy of cluster updates: – Node by node – Environment by environment – Verification process.

David Foster, LCG For PEB Meeting 9/4/02 People and Activities • March 2002. Hearns: David Foster, LCG For PEB Meeting 9/4/02 People and Activities • March 2002. Hearns: Working on XML configuration description and interpretation. • Mid-April 2002, Neilson: Production service manager setting up “stable testbed” • Mid-April 2002, Tomlin: Alarm management handling, monitoring system integration. • June 2002, TBD: Updaterpms solution for package distribution • Now: Maite: High level description of system configuration. • Other Activities: Kickstart file generation from configuration database, Scalable access to RPM’s

David Foster, LCG For PEB Meeting 9/4/02 Fabric Metrics • Need to understand how David Foster, LCG For PEB Meeting 9/4/02 Fabric Metrics • Need to understand how the fabric should perform CPU -> Disk Storage -> Mass Storage • CDR (Write) • Bulk Reprocessing Events (Read) • Analysis (Read/Write) Need some Fabric Data Challenges (Panzer)

David Foster, LCG For PEB Meeting 9/4/02 CDR • Mainly Alice Data Challenges • David Foster, LCG For PEB Meeting 9/4/02 CDR • Mainly Alice Data Challenges • CPU -> mass storage end to end 200 MB/Sec in 2002 300 MB/Sec in 2003 450 MB/Sec in 2004 750 MB/Sec in 2005

David Foster, LCG For PEB Meeting 9/4/02 Reprocessing Events • Understand how fabric architecture David Foster, LCG For PEB Meeting 9/4/02 Reprocessing Events • Understand how fabric architecture affects throughput. – CPU/Disk/Tape server organisation • Determine reasonable goals (MB/Sec, events/unit time)

David Foster, LCG For PEB Meeting 9/4/02 Analysis Many open issues …. . • David Foster, LCG For PEB Meeting 9/4/02 Analysis Many open issues …. . • Data Organisation/Migration (across the grid) • Physical/Logical structure (across servers and disks) • Reliability/Scalability (fault tolerence and access patterns)

David Foster, LCG For PEB Meeting 9/4/02 Actions • Establish Fabric Metrics – CDR David Foster, LCG For PEB Meeting 9/4/02 Actions • Establish Fabric Metrics – CDR and data reprocessing – Fabric Computing/Data Challenges • Start simple modelling of data organisation and fabric structure (Post Pasta III ? ) – Make some simple assumptions about object sizes, access patterns etc. – Understand bottlenecks and architecture issues. – Understand failure modes and fault management

David Foster, LCG For PEB Meeting 9/4/02 Summary Date LCG Milestones FM Milestones Fabric David Foster, LCG For PEB Meeting 9/4/02 Summary Date LCG Milestones FM Milestones Fabric Milestones CDR Target June 30 ‘ 02 Production testbed created Initial System 20 nodes ? Sept 30 ‘ 02 Pasta III Completed Semi. Automated 200 nodes Detailed Fabric Model Started Nov 30 ‘ 02 Glue Final Recommendations 200 MB/sec Jan 01 ‘ 03 LCG Software Integration starts Mar 31 ‘ 03 LCG Production 1. 0 Fully Automated 400 nodes ? Fabric Modeling Completed Dec 31 ‘ 03 EDG Final Deliverable 800 nodes 400 MB/sec