Practical Parallel Processing for Realistic Rendering
T. Michael Davis, Alan G. Chalmers, Henrik Wann Jensen · 2000
definition of a task . ................... 35 3.1.4 System architecture . ....................... 36 3.2 ComputationalModels ........................... 37 3.2.1 Data driven model . . ....................... 38 3.2.2 Demand driven model ....................... 42 3.2.3 Hybrid computational model ................... 47 3.3 TaskManagement ............................. 47 3.3.1 Task definition and granularity . . . . . . ............ 47 3.3.2 Task distribution and control . ................... 49 3.3.3 Algorithmic dependencies . . ................... 49 3.4 TaskSchedulingStrategies ......................... 53 3.4.1 Data driven task management strategies . . ............ 53 3.4.2 Demand driven task management strategies ............ 53 3.4.3 Task manager process ....................... 57 3 3.4.4 Distributed task management ................... 59 3.4.5 Preferred bias task allocation ................... 61 4 Data Management 64 4.1 WorldModeloftheData:NoDataManagementRequired ........ 64 4.2 VirtualSharedMemory .......................... 65 4.2.1 Implementing virtual shared memory . . . ............ 65 4.3 TheDataManager ............................. 66 4.3.1 The local data cache . ....................... 67 4.3.2 Requesting data items ....................... 68 4.3.3 Locating data items . ....................... 70 4.4 Consistency................................. 74 4.4.1 Keeping the data items consistent . . . . . ............ 74 4.4.2 Weak consistency: repair consistency on request . . . . ..... 77 4.4.3 Repair consistency on synchronisation: Release consistency . . . 77 4.5 Minimising the Impact of Remote Data Requests . ............ 77 4.5.1 Prefetching . . . . . . ....................... 78 4.5.2 Multi-threading . . . .........................