![(please configure the [header_logo] section in trac.ini)](/trac/PP2009/chrome/site/your_project_logo.png){kind=logo}
Changes between Version 5 and Version 6 of PTVS_Intro
- Timestamp:
- Dec 4, 2009, 5:54:53 PM (14 years ago)
Legend:
- Unmodified
- Added
- Removed
- Modified
-
PTVS_Intro
v5 v6 6 6 '''''What has happened? What's been going on?''''' 7 7 8 The adoption of simulation techniques for the analysis of large-scale, dynamic and complex vehicular networks and related wireless communication-based services, creates severe problems in terms of simulation resources and simulation performance. 8 The adoption of simulation techniques for the analysis of large-scale, dynamic and complex vehicular networks 9 and related wireless communication-based services, creates severe problems in terms of simulation resources and 10 simulation performance. 9 11 10 In a simulation, the individual evolution of a massive number of entities, the dynamic correlation effects due to mobility and the local broadcast nature of the wireless signal propagation effects in urban scenarios greatly increase computational requirements. 12 In a simulation, the individual evolution of a massive number of entities, the dynamic correlation effects due 13 to mobility and the local broadcast nature of the wireless signal propagation effects in urban scenarios greatly 14 increase computational requirements. 11 15 12 A sequential simulation of a real scenario with a large number of complex model entities (like vehicles running distributed services and their wireless communications) may become computationally infeasible on a single physical execution unit. This is why solutions based on parallel computing architectures like SIMD architectures have been considered, but they are usefull only for homogeneous models. 16 A sequential simulation of a real scenario with a large number of complex model entities (like vehicles running 17 distributed services and their wireless communications) may become computationally infeasible on a single physi- 18 cal execution unit. This is why solutions based on parallel computing architectures like SIMD architectures have 19 been considered, but they are useful only for homogeneous models. 13 20 14 Parallel and HPC architectures have high cost and require expertise in parallel programming to exploit the architecture potential. The Parallel and Distributed Simulation (PADS) exploit concurent computation over different execution units. 21 Parallel and HPC architectures have high cost and require expertise in parallel programming to exploit the archi- 22 tecture potential. The Parallel and Distributed Simulation (PADS) exploit concurent computation over different 23 execution units. 15 24 16 25 == So What? == … … 18 27 '''''What should we do? Why this?''''' 19 28 20 Single-threaded simulation engines based on instantaneous events have just one current event. In contrast, multi-threaded simulation engines supporting an interval based event model may have multiple current events. In both cases there are significant problems with synchronization between current events. 29 Single-threaded simulation engines based on instantaneous events have just one current event. In contrast, multi- 30 threaded simulation engines supporting an interval based event model may have multiple current events. In both 31 cases there are significant problems with synchronization between current events. 21 32 22 VNSim is a discrete event simulator, which means that the simulation time advances with a fixed time resolution after executing the application code for the current coment of the simulation time. 23 This means that at every moment of the simulation time, all the current events are pulled from the queue of events and handeled in a random order. 33 VNSim is a discrete event simulator, which means that the simulation time advances with a fixed time resolution 34 after executing the application code for the current comment of the simulation time. This means that at every 35 moment of the simulation time, all the current events are pulled from the queue of events and handeled in a random 36 order. 24 37 25 In event based simulators time increases variably, based on the occurence of events. 26 When the engine advances the simulation time, it pulls out from the main event queue that it manages all the events for the new moment of time. 38 In event based simulators time increases variably, based on the occurrence of events. 39 When the engine advances the simulation time, it pulls out from the main event queue that it manages all the events 40 for the new moment of time. 27 41 28 We want to find a solution for using multi-threading processing in order to obtain increased performances and speadup in the event-queue processing. 29 First we split the map in independent regions. We want to keep multiple event-queues and in every queue we hold only events associated to a certain region of the map, independent from others regions. Based on the location where the event was produced, this event is introduced in a certain event-queue. Different thread will process different event-queues. 42 We want to find a solution for using multi-threading processing in order to obtain increased performances and 43 speedup in the event-queue processing. First we split the map in independent regions. We want to keep multiple event 44 -queues and in every queue we hold only events associated to a certain region of the map, independent from others 45 regions. Based on the location where the event was produced, this event is introduced in a certain event-queue. 46 Different thread will process different event-queues. 30 47 31 What we need is a priority queue algorithm. The pending event queue is organized as a priority queue sorted by event time. This means that regardless of the order in which events are added, they are removed in strictly chronological order. Events are then scheduled dynamically as the simulation proceeds. 48 What we need is a priority queue algorithm. The pending event queue is organized as a priority queue sorted by event 49 time. This means that regardless of the order in which events are added, they are removed in strictly chronological 50 order. Events are then scheduled dynamically as the simulation proceeds. 32 51 33 Every event has a reference to a thread who will process it, and when that event is pulled off from the queue that thread is restarted/ awaken. So the events comming from independent intersections will be processed in parallel, because they are associated to different threads. 52 Every event has a reference to a thread who will process it, and when that event is pulled off from the queue that 53 thread is restarted/ awaken. So the events coming from independent intersections will be processed in parallel, 54 because they are associated to different threads. 34 55 35 56 A simulation engine may have the following logic: … … 39 60 START:[[BR]] 40 61 41 -initialize ending condition to false[[BR]] 42 43 -initialize system state variables[[BR]] 44 45 -initialize clock (usually starts at simulation time zero)[[BR]] 46 47 -schedule an initial event (put some initial event into the event queue)[[BR]] 62 * initialize ending condition to false[[BR]] 63 * initialize system state variables[[BR]] 64 * initialize clock (usually starts at simulation time zero)[[BR]] 65 * schedule an initial event (put some initial event into the event queue)[[BR]] 48 66 49 67 DO LOOP or WHILE LOOP:[[BR]] … … 51 69 While (ending condition is false) do[[BR]] 52 70 53 -set clock to next event time[[BR]] 54 55 -process next event and remove from the event-queue[[BR]] 56 57 -update statistics[[BR]] 71 * set clock to next event time[[BR]] 72 * process next event and remove from the event-queue[[BR]] 73 * update statistics[[BR]] 58 74 59 75 END:[[BR]] 60 76 61 -generate statistics[[BR]]77 * generate statistics[[BR]] 62 78 63 79
