Parallel simulation of complex systems

该方案包括两个部分：（1）分布式Verilog仿真（2）并行粒子仿真。分布式Verilog仿真的研究是一个现有的项目，其目标是开发一个可扩展的分布式门级Verilog仿真器的分布式和共享内存平台的延续。 我们以前的研究已经导致了一个分布式的Verilog仿真环境，DVS，使用时间扭曲作为仿真引擎和开源的Icarus Verilog作为前端的发展。 我们已经开发了一个高效的门级模拟器，XTW，门级模拟器，它利用了多级并行机制。 我们建议（1）开发一个仿真环境，通过Time Warp将在各个节点上执行的Modelsim顺序仿真器链接在一起（2）将XTW移植到共享内存环境中。我们将尝试利用共享内存来改进内存管理。(3)继续我们在基于模块的负载平衡方面的工作。初步结果表明，基于模块的划分受益于局部性。我们提出了在仿真执行过程中重新划分电路的有效算法的发展（4）将强化学习技术应用于乐观仿真。 我们的第二个研究领域集中于在宇宙学模拟器的背景下，为求解偏微分方程和离散事件算法的连续方法的集成。 我们正在使用一个标准的技术解决相互作用（宇宙学）机构-牛顿引力方程的oct树求解器。我们的目标是通过使用离散事件技术将碰撞物体纳入图片中。我们已经实现了一个时间扭曲引擎，并将其与八叉树求解器集成。这项工作是与华盛顿大学的一个天体物理小组合作完成的。我们提出的第一个任务是验证的组合算法和评估其可扩展性，使用现实的情况下提供的天体物理学家。 然后我们打算开发分区算法。

There are two components of this proposal (1) distributed Verilog simulation (2) parallel particle simulatlon. The proposed research on distributed Verilog simulation is a continuation of an existing project the goal of which is to develop a scalable distributed gate level Verilog simulator for both distributed and shared memory platforms. Our previous research has resulted in the development of a distributed Verilog simulation environment, DVS, using Time Warp as the simulation engine and the open source Icarus Verilog as a front end. We have developed a highly efficient gate level simulator, XTW, for a gate level simulator which makes use of a multi-level queueing mechanism. We propose (1) developing a simulation environment which links Modelsim sequential simulators executing on individual nodes together via Time Warp (2) porting XTW to a shared memory environment. We will try to take advantage of shared memory for improved memory management. (3) continuation of our work on module based load balancing. Preliminary results indicate that partitioning based upon modules has benfited from the locality. We propose the development of efficient algorithms for re-partitioining the circuit during the course of the simulation's execution (4) applying reinforcement learning techniques to optimistic simulations. Our second research area focuses on the integration of continuous methods for the solution of pde's and discrete event algorithms in the context of cosmological simulators. We are making use of a standard technique for the solution of interacting (cosmological) bodies-an oct tree solver for Newtonian gravitational equations. Our goal is to incorporate colliding bodies into the picture by making use of discrete event techniques. We have implemented a Time Warp engine and integrated it with the oct-tree solver. The work is being done in conjunction with an astro-physics group from the University of Washington. Our first proposed task is the verification of the combined algorithms and an evaluation of their scalability using realistic scenarios provided by the astrophysicists. We then intend to develop partitioning algorithms.