Changes between Version 33 and Version 34 of ParaView
- Timestamp:
- 05/03/17 16:59:56 (7 years ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
ParaView
v33 v34 55 55 * Start !ParaView and connect to the pvservers (localhost:11111). 56 56 57 ==== How to read data to !ParaView in parallel 58 Just starting !ParaView in parallel __does not__ result necessarily in using the compute resources (cpu and memory) of more than one node or using them on the allocated nodes equaly. 57 ==== How to use !ParaView in parallel 58 Just starting !ParaView in parallel __does not__ result necessarily mean to benefit from the compute resources (cpu and memory) of more than one node.\\ 59 Even if !ParaView distributes the data and work over the compute nodes this might not happen equally.\\ 59 60 60 61 Two main issues must be considered: 61 * Parallel Data Management - the data must be distributed across parallel processes to take advantage of resources 62 * Parallel Work Managemnet - the work must be distributed across parallel processes to take advantage of 62 * Parallel Data Management - the data must be distributed equally across parallel processes to take advantage of resources 63 * Parallel Work Management - the work must be distributed equally across parallel processes to take advantage of resources 64 65 You have less influence on the 'Parallel Work Management', therefore we only discuss 'Parallel Data Management' for now. 63 66 64 67 ===== Parallel Data Management 65 68 Data must be distributed across parallel processes to take advantage of resources.\\ 69 This distribution can be accomplished by the reader or the D3 filter afterwards. 70 71 * fully parallel readers 72 * Explicit parallel formats use separate files for partitions (.pvti, global.silo) 73 * Implicit parallel formats – parallel processes figure out what they need (.vti, brick-f-values) 66 74 67 68 * Some !ParaView readers import in parallel 69 * Explicit parallel formats use separate files for partitions (.pvti, global.silo) 70 * Implicit parallel formats – parallel processes figure out what they need – (.vti, brick-f-values) 71 * Some !ParaView readers may seem to import in parallel 72 * Actually, import serially and then distribute 73 * Bad bad bad – 1 process needs enough memory for entire dataset plus additional space for partitioning 74 * Some !ParaView readers do NOT read in parallel 75 * ... and leave it to you (D3 filter in Paraview - this results in an unstructured grid, which might need more memory) 76 * See Bad bad bad above 75 * serial readers + distribute 76 * first process needs enough memory for __entire__ dataset plus additional space for partitioning 77 77 78 ====== Test Parallel Data Mangement 79 * Click Sources->Sphere 80 * Max out Theta Resolution and Phi Resolution 81 * Click Filters->Alphabetical->Process Id Scalars 82 * => Segments are colored by which process handles them 78 * fully serial readers 79 * you need to distribute data manually using D3 filter 80 * attention: D3 filter outputs unstructured grid, which might need more memory 81 82 ===== Test Parallel Data Mangement 83 * Load your data 84 * Add filter 'Process Id Scalars' 85 * Segments are colored by which process handles them 83 86 84 87 ----