四舍五入如何提高性能(如果有的话)? [英] How (if at all) could rounding Decimals improve performance?

问题描述

我很好奇，当我在Spark程序中对一堆小数进行任何形式的数学运算时，它对舍入数字有什么影响吗?例如:

I am curious when I am doing any sort of math on a a bunch of decimals in my Spark program does it make any difference to round numbers beforehand? For instance does:

3.000009481 * 10.90001001010

3.000009481 * 10.90001001010

计算与以下内容相同的时间:

Calculate the same time as something like:

3.000 * 10.900

3.000 * 10.900

在乘法，除法以及幂和平方根等更高级的内容上是否有所不同?

Does it differ for multiplication, division, and more advanced things like power and square root etc?

我想知道，因为当您对数百万个数字进行处理时，可以说它是毫秒级的整数吗?还是总是用小数点获得相同的计算时间?

I am wondering because lets say its a fraction of a millisecond when you are doing it on millions of numbers does it make sense to round? Or will you always get the same calculation times with decimals?

我有一个很大的aggregateByKey函数，正在执行大量计算.我所在的群集始终提供不一致的性能指标.

I have a large aggregateByKey function that is doing a lot of computation. The cluster I am on consistently provides inconsistent performance metrics.

推荐答案

从数字上看，它可以产生很大的不同(请参阅:

It can make a huge difference numerically (see: What Every Computer Scientist Should Know About Floating-Point Arithmetic) but won't affect performance.

这篇关于四舍五入如何提高性能(如果有的话)?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！