Carothers, C. D., K. S. Perumalla, and R. M. Fujimoto. 1999(b). E- cient optimistic parallel simulations using reverse computation," (journal version). ACM Transactions on Computer Modeling and Simulation (TOMACS), 9(3): 224-253.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....state, only the modified parts of a state are recorded. However, in programs where the modified state space is large, memory and time overheads of incremental state saving might again exceed a#ordable limits. Carothers et al. introduce another approach for optimistic parallel simulations [7]. This approach is source transformation. In source transformation, the source code (e.g. in C) is transformed to a reversible source code version excluding destructive statements such as direct assignments. For destructive statements, state saving is applied. Consequently, the execution time and ....

C. Carothers, K. Perumalla, and R. Fujimoto. E#cient optimistic parallel simulations using reverse computation. ACM Transactions on Modeling and Computer Simulation, 9(3), July 1999.

....using native machine instructions, not even in the forward direction. Moreover, since reversible instructions are usually constructed as stack operations, a significant amount of stack memory may be required in program animation. Another approach introduced is the source transformation approach [4]. In source transformation, the source code (e.g. in C) is transformed to a reversible source code version excluding destructive statements such as direct assignments. For destructive statements, state saving is applied. Consequently, the execution time and memory requirement of the transformed ....

....Table 1: The sizes of the benchmarks. #C lines 12 16 18 14 #assembly instructions 15 37 59 35 Tables 2 and 3 show memory and time overhead results of the RCG algorithm, the ordinary incremental state saving (ISS) 11] and incremental state saving for only destructive instructions (ISSDI) [4]. The memory overhead measurements were performed in the following way: For ISS and ISSDI, we calculated the program points where state saving is needed with each benchmark and we instrumented each benchmark with memory store instructions which save state at the calculated points. Then, we applied ....

C. Carothers, K. Perumalla, and R. Fujimoto. E#cient optimistic parallel simulations using reverse computation. ACM Transactions on Modeling and Computer Simulation, 9(3), July 1999.

....machine instructions directly, not even in the forward direction. Moreover, since reversible instructions are usually constructed as stack operations, a significant amount of stack memory may be required in program animation. Another approach introduced is the source transformation approach [5]. In source transformation, the source code (e.g. in C) is transformed to a reversible source code version excluding destructive statements such as direct assignments. For destructive statements, state saving is applied. Consequently, the execution time and memory requirement of the transformed ....

.... is connected to the PowerPC board via a background debug mode interface [8] Figures 8 and 9 show memory and time overhead comparisons, respectively, between the RCG algorithm, the ordinary incremental state saving (ISS) 10] and incremental state saving for only destructive instructions (ISSDI) [5]. The benchmark programs used are a Fibonacci number generator (FNG) with 100 iterations, a selection sort (SS) with 10 inputs, a 3 by 3 matrix multiplication (MM) and a random number generator (RNG) with 100 iterations. The results indicate that RCG algorithm achieves from 3.17X to 400X and from ....

C. Carothers, K. Perumalla, and R. Fujimoto. E#cient optimistic parallel simulations using reverse computation. ACM Transactions on Modeling and Computer Simulation, 9(3), July 1999.

....LGA models with as few as two bits of di#using state at each lattice site can be constructed using the dimensionsplitting technique. The benefits of similar economies in simulation state have recently been discussed in other contexts. For example, in the context of reversible compilers see Ref. [35]; in the context of systems relying on pseudorandom numbers see Ref. 36] It might seem that the operation on one dimension at a time would result in slower simulations than a more complex LGA model, which mixes data in several dimensions at once. However, from a computational complexity ....

C. Carothers, K. Perumalla, and R. Fujimoto. E#cient optimistic parallel simulations using reverse computation. In ACM Transactions on Modeling and Computer Simulation, volume 9, 1999.

No context found.

Carothers, C. D., K. S. Perumalla, and R. M. Fujimoto. 1999(b). E- cient optimistic parallel simulations using reverse computation," (journal version). ACM Transactions on Computer Modeling and Simulation (TOMACS), 9(3): 224-253.

No context found.

C. D. Carothers, K. S. Perumalla, R. M. Fujimoto. E#cient optimistic parallel simulations using reverse computation. ACM TOMACS, 9(3):224--253, 1999.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC