Parallel Computing

The three major concerns of industrial CAE users are accuracy, computation speed and user-friendliness. True 3D simulation not only satisfies those concerns, but offers more advantages that conventional 2.5D cannot reach, such as CAD integration, accuracy, minimized model simplification...etc. However, true 3D simulation inevitably increases computation time and requests more memory. Although the High-Performance Finite Volume Method, HPFVM, employed by Moldex3D/Solid has already outperformed other 3D software, but users are still eagerly expecting significant improvements.

For optical parts, fiber-reinforced automobile components, connectors, gears...etc., the demand for high accuracy and high speed computation can never be underestimated. Computation speed can be improved by newer and more powerful CPU. However, improvement from CPU clock rate alone can not satisfy industrial users in speed nor accuracy. Utilization of multiple CPU is therefore the most effective approach.

Parallel computing is fast becoming an inexpensive alternative to standard supercomputer for solving large scale problems that arise in scientific and engineering applications. Generally, there are two types of parallel computing platform: (1)Symmetric Multiple Processor, SMP, and (2) Massively Parallel Processing, MPP. The CPUs of a SMP platform share the same memory and are controlled by a single operating system. Common SMP computers are supercomputer, dual CPU server, 4-CPU server...etc. A lot of industries have already used SMP computers as servers for their simplicity and easy maintenance. MPP platform consists of multiple computers. Each computer has its own CPU and operating system. The communication and collaboration between computers are done through high-speed network (Myrinet or Gigabit ethernet) and certain message interfaces. Common MPP platforms consist of a cluster of standard PC or high-end workstation. By utilizing standard PCs, PC cluster generally offers the best cost/performance. The cost in operation system can be even lower if Linux rather than Windows is employed. Supercomputer has a lot of advantages, but it's also too expensive to afford for most industies. PC cluster has the advantages of high performance and low-cost, so it's usually referred as "the Poor Man's Supercomputer" Among the 500 most powerful computer systems in the world, many are PC clusters. (www.top500.org)

Now Moldex3D/Solid takes the lead in parallel computing to enhance the computation performance and perform analyses in less time on more complex model with larger element count than ever. The high-performance parallelized kernel is equipped integrated analyses of Flow, Pack, Cool, Warp, Fiber Orientation and RIM (Reactive Injection Molding). Furthermore, Moldex3D/Solid parallel computing technology supports both Multi-CPU platform and PC-Cluster.

The most common SMP computers are dual CPU desktop PC, which is only 20~35% more expensive than single CPU one. With dual-CPU power, the computation speed can typically have 60% to 75% acceleration. Tetra (4) and Octa (8) CPU high-end servers are also available from major computer suppliers. Users can gain significant performance enhancement by slightly increase their hardware investment. Besides, PC cluster is proven to be much more impressive in terms of cost-performance. In general, 4-node cluster is about 250~300% faster than a single CPU PC. Twenty times speedup is possible by using a 32-node PC cluster.

Performance Benchmark
The following table demonstrates the parallel computing performance of Moldex3D/Solid-Flow with various models. The testing platform is Dual Intel Xeon 2.4GHz CPU and Windows XP Professional. For Moldex3D/Solid-Flow, although speedup partly depends on the geometry, it can typically have 60% to 75% acceleration. Flow analysis is usually the most time-consuming step in all analyses. This improvement implies a big reduction in overall analysis time.

	Platform:Dual Intel Xeron 2.4GHz CPU with 2 GB RAM System:Windows XP Professional Solver:Moldex3D/Solid-Flow R7.0 Case# Elements 1 CPU(sec) 2 CPU(sec) *Speed Up Case1 7,586 33 23 1.43 Case2 113,978 8,400 5,300 1.58 Case3 560,716 74,100 42,300 1.75   Note:Speed UP:Time(1 CPU)/Time(2 CPU)

The following table shows the parallel computing performance of Moldex3D/Solid-Warp. Generally, it has 70% - 88% acceleration.

	Platform:Dual Intel Xeron 2.4GHz CPU with 2 GB RAM System:Windows XP Professional Solver:Moldex3D/Solid-Flow R7.0 Case# Elements Nodes 1 CPU(sec) 2 CPU(sec) Speed Up Case1 12,500 15,000 119 64 1.86 Case2 125,421 37,026 71 41 1.73 Case3 128,456 49,258 98 54 1.81 Case4 571,392 107,516 521 277 1.88 Case5 1,233,203 291,784 2,721 1,688 1.61   Note:Speed UP:Time(1 CPU)/Time(2 CPU)

Except to accelerate the computation speed, another big advantage of parallel computing is the capability to deal with the huge models. Current 32-bit CPU can only address up to 4GB RAM. Excluding the memory reserved for Windows itself, an application program can only access up to 3GB memory. Therefore, memory may not be enough for some big scale problems. Moldex3D parallel computing makes that it is possible for 32-bit CPU to calculate bigger models than ever. For example, the following table demonstrates the application of Dual Xeon CPU to a huge model with 3.2 million elements. Although the machine is equipped with 8GB RAM, memory can not be applied enough when it is on a single CPU due to the limitation of CPU itself. Hence, using Moldex3D parallel computing, it comes up to pass this limitation.

	CPU Intel Xeron 2.8GHz Mesh 3.2 Million elements CPU number 1 2 RAM 8.0GB Operating System Windows 2003 Server Enterprise 32 bit Solid/Flow R7.0 Memory Not Enough 24.4 hour

System Requirements :

The parallel computing of Moldex3D requires distributed systems with high performance interconnects. The amount of data passed between processors in a typical Moldex3D run could be in hundreds to thousands megabytes. The followings are the basic system requirements.

Managing Node:
1. Microsoft Windows 2003 server recommended.
2. Intel Pentium, Intel Xeon, Intel EM64T, AMD Athlon, or AMD Opteron based processor.
3. 2.0 GB RAM or greater.
4. Gigabit Ethernet or greater

Computing Node:
1. Microsoft Windows XP Professional, Windows XP x64, or Windows 2000 recommended.
2. Intel Pentium, Intel Xeon, Intel EM64T, AMD Athlon, or AMD Opteron based processor.
3. 1.0 GB RAM or greater.
4. Gigabit Ethernet or greater.

Network:
1. Gigabit switch or greater.