Visual Solution to High Performance Computing Computer and Automation Research Institute Laboratory of Parallel and Distributed Systems
The Big Issue Sequential program code C, C++ Fortran Parallel program code ? Parallelization? Debugging? Performance? Execution?
Distributed supercomputing with P-GRADE Parallel Grid Run-time and Application Development Environment (P-Grade) The first highly integrated parallel application development system for Grid and clusters Provides: –Parallel, supercomputing programming –Fast and efficient development –Observation and visualization –Fault and performance analysis ON CLUSTERS ON SUPERCOMPUTERS IN THE GRID
DESIGN: Three hierarchical layers for easy parallelisation Application window Process window Text window
DESIGN: Built-in Communication Templates for fast design
DEBUGGING: Hierarchical control and views to eliminite programming errors
DEBUGGING: Macrostep-by-macrostep… Support for systematic debugging to handle non- deterministic behaviour of parallel applications Automatic dead-lock detection Replay technique with collective breakpoints Active control: generation of Execution Trees Testing parallel programs by temporal logic assertions
MONITORING: statistics and event traces for performance evaluation Collection of both statistics and event traces Application monitoring and visualization in the Grid Evaluation of long-running programs based on semi- on-line trace collection Support for debugger in P-GRADE by execution visualisation
PERFORMANCE VISUALIZATION: Visualization of Event Traces User controlled focus on processors, processes and messages
P-GRADE: Stages of Software Development and Execution Execution Design, debugging Performance- analyzis Grid & Cluster & Supercomputer
Application I: Ultra-short range weather forecast (Hungarian Meteorology Service) 25 x 10 x25 x5 x Forecasting dangerous weather situations (storms, fog, etc.), crucial task in the protection of life and property Processed information: surface level measurements, high-altitude measurements, radar, satellite, lightning, results of previous computed models Requirements: Execution time < 10 min Higher resolution Result
Processing of satellite images Transformation Interpolation to basic GRID Processing: Altitude of clouds & Satellite raw image Receiving image
P-GRADE GUI DB P-GRADE in meteorology demonstrated at Euro-Par 2003 conference Hungarian Meteorological Service P-GRADE - RS CONDOR-PVM Parallel execution and Monitoring TRACE & Results Jobsubmission Input
Workflow support (in the Grid) Illustrated using the meteorological application Job Secure data transfer Input files Monitorin g of Workflow Monitorin g of Job
APPLICATION II: Urban traffic simulation (University of Westminster)
Checkpointing and migration of P-GRADE programs between clusters Budapest, MTA SZTAKI London-UoW 4 P-GRADE Program Runs at UoW cluster 3 P-GRADE program migrates to London as a Condor job SZTAKI & BUTE Clusters overloaded CHECKPOINTING SZTAKI & BUTE P-GRADE GUI P-GRADE program Submitted to Budapest as a Condor job 1 2 P-GRADE program Runs at SZTAKI cluster
Monitoring & visualisation - job migration in the Grid - Slow-down of simulator job due to overloaded cluster Migration of job between clusters in the GRID
Summary: Advantages of P-GRADE Effective and efficient support for each stage of the parallel program development Fast parallelization of existing algorithms Reusability of sequential code Hiding of low level communication functions Integrated graphical concept Predefined communication templates Support for hierarchical design Even non-professional programmers can use it (without a steep learning curve) Portability (supercomputers, PC clusters, Grid)
Parallel Grid Run-time and Application Development Environment