Printable View
Hi all,
Linpack has become the most popular high-performance computer system for testing floating-point performance benchmark. Through the use of high-performance computers, I a using Gaussian elimination method for solving the test, but i want to know how to run LinPack stress test. Please suggest.
First you go process mapping. Adjustment process and the mapping between processor performance produced no small effect, optimization of this mapping is the key to change the calculation load of each node and communication operator to reduce the competitive communications networks, to achieve faster communication path and the realization of nodes Load Balancing. Such as: the computation load is too focused on a particular number of nodes between two nodes at the same time to avoid multiple concurrent communication on the process.
The LinPack decomposition process is a relatively time-consuming process, in order to improve performance, its use as part of the first decomposition, and then broadcast out of this part of the results. "DEPTHs"Value is the instructions LBroadcast several times.DEPTHs＝0 that would LOne-time radio out, that is, the whole LDecomposition is completed in a one-time radio;DEPTHs＝1that would LBroadcast twice; and so forth.
Using LinPack compiler optimization can greatly improve CPU Performance-intensive applications, such as the use of very long instruction can be greatly enhanced floating-point processing performance. Compiler optimization under the condition without modifying the program There are two main methods: use a better compiler performance and increase the compiler optimization options. InX86Platform, the main compiler has GNU、Intel Compiler,PGI Compilers, in some specific platforms have a special compiler, such asIBM Of PSeries machine has its proprietary compilerxlc Andxlf. Compiler compiler optimization options and is closely related to the different compiler optimization options are different.
Linpack tests include three types, Linpack100, Linpack1000 and HPL. Linpack100 size of 100 bands for solving dense linear algebraic equations, which only allows to optimize the compiler optimization options, may not change the code, or even comments in the code not to modify. Linpack1000 required size of 1000 Order of solving linear algebraic equations, to achieve the specified accuracy, without changing the premise of doing computational algorithm and code optimization to do.
High Performance Linpack, also known as highly parallel computing benchmark, its array of size N does not limit the scale of the problem solving can be changed, in addition to the basic algorithm (computing) immutable, the can use any other optimization methods. The first two tests runs a smaller scale, is not very suitable for the development of modern computers, so more tests now use standard HPL, and the order of N is linpack test parameters must be specified.
The optimization is not done once to often need to cycle, because with the new application to join the new node expansion, increased accuracy, model complexity increased, emergence of new theories and algorithms and the use of new hardware, such as SSD, etc. , requires the application of continuous testing and performance optimization, the only way to ensure the system is always in the best state
The LinPack stress test is an amazingly well-optimized app that solves a system of simultaneous linear equations by the Gaussian elimination method.LinX is a simple linpack interface that allows you to easily test and stress your CPU.It stresses the CPU much much more than Prime95,Orthos and can save you lots of time.