当前位置:网站首页>Rust : 性能测试criterion库
Rust : 性能测试criterion库
2020-11-08 11:21:00 【osc_ju8o7gji】
在量化领域,对性能要求比较高,特别是高频交易,那是纳秒必争。在RUST中,测试一个函数,或一个操作耗时即性能分析,应是如何做呢?
一、计时器
是否可以用std::time::systime 来计算花时情况?我们来试一试:
use std::time::SystemTime;
pub struct Stock{
pub price:f64,
pub volume:i32,
pub code:String,
pub amount:f64,
}
impl Stock{
fn default()->Self{
let p =10000.0_f64;
let v =1_i32;
Stock{
price:p,
volume:v,
code:String::from("SH600036"),
amount: p*v as f64,
}
}
}
fn main() {
let now = SystemTime::now();
let stock = Stock::default();
let elapsed = now.elapsed().unwrap().as_nanos();
println!("cost time :{:?} ",elapsed);
}
你会发现:
为什么创建一个结构体要花3500纳秒,这也太不可思议了吧。其实,真实的情况并不是这样的,是因为计时器的测试误差导致。
所以,如果要进行专业的性能测试,还是要借助于专业三方性能测试库。
二、专业三方库
RUST中,专业三方库还是有不少,推库用criterion.rs.
https://github.com/bheisler/criterion.rs
具体大家可以看看。
三、实例
1、创建测试目录
在工程中,建一个专门的测试用的目录:我这儿是在
my_bench目录下,专门创建了benches目录
2、toml文件
对于my_bench工程下toml文件
toml文件中,增加:
[dev-dependencies]
criterion = "0.3"
[[bench]]
name = "my_benchmark"
harness = false
其中,name ="my_benchmark"是有所指的,不是随便写的。这里是指,我在my_bench工程下,有一个my_benchmark.rs文件,里面有我写的性能测试代码。别的地方就不要去找了。
3、准备性能测试代码
因为toml文件中,有name =“my_benchmark” ,那么,性能测试代码就是my_benchmark.rs. 创建这个文件。
写入相应的性能测试代码:
use criterion::{
black_box, criterion_group, criterion_main, Criterion};
pub struct Stock{
pub price:f64,
pub volume:i32,
pub code:String,
pub amount:f64,
}
impl Stock{
fn default()->Self{
let p =10000.0_f64;
let v =1_i32;
Stock{
price:p,
volume:v,
code:String::from("SH600036"),
amount: p*v as f64,
}
}
}
pub fn set_heap()->Stock{
Stock::default()
}
pub fn box_stock() ->Box<Stock>{
Box::new(Stock::default())
}
fn criterion_benchmark_heap(c: &mut Criterion) {
c.bench_function("stock ", |b| b.iter(|| set_heap()));
}
fn criterion_benchmark_box(c: &mut Criterion) {
c.bench_function("box<stock> ", |b| b.iter(|| box_stock()));
}
criterion_group!(benches, criterion_benchmark_heap,criterion_benchmark_box);
criterion_main!(benches);
4、运行
在所有的工程my_bench下,运行cargo bench,即可以进行性能测试代码的运行了。
Finished bench [optimized] target(s) in 2.80s
Running target\release\deps\my_bench-95230ab505784caf.exe
running 0 tests
test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out
Running target\release\deps\my_benchmark-a468db13a6ba0ea3.exe
Gnuplot not found, using plotters backend
stock time: [75.387 ns 83.206 ns 91.985 ns]
Found 3 outliers among 100 measurements (3.00%)
3 (3.00%) high mild
box<stock> time: [168.68 ns 189.43 ns 212.00 ns]
Found 3 outliers among 100 measurements (3.00%)
从上面,我们可以得到详细的测试信息,包括运行的时间分布情况。
专业性能测试工具表明,创建一个Stock对象,大约平均需要73ns,而不是计时器显示的3500ns;
但是,创建一个Stock 的Box对象需要的时间是双倍。
版权声明
本文为[osc_ju8o7gji]所创,转载请带上原文链接,感谢
https://my.oschina.net/u/4288530/blog/4708045
边栏推荐
猜你喜欢
Installing MacOS 11 Big Sur in virtual machine
How does spotify drive data-driven decision making?
i5 1135g7和i5 1035g1参数对比区别大吗? 哪个好
笔试面试题目:判断单链表是否有环
PX4添加新的应用
2018中国云厂商TOP5:阿里云、腾讯云、AWS、电信、联通 ...
运维人员常用到的 11 款服务器监控工具
Analysis of ArrayList source code
比Python快20%,就问你兴不兴奋?
It's 20% faster than python. Are you excited?
随机推荐
211考研失败后,熬夜了两个月拿下字节offer!【面经分享】
Mate 40 series launch with Huawei sports health service to bring healthy digital life
next.js实现服务端缓存
2018中国云厂商TOP5:阿里云、腾讯云、AWS、电信、联通 ...
Top 5 Chinese cloud manufacturers in 2018: Alibaba cloud, Tencent cloud, AWS, telecom, Unicom
个人目前技术栈
Xamarin 从零开始部署 iOS 上的 Walterlv.CloudKeyboard 应用
Second assignment
阿里出品!视觉计算开发者系列手册(附网盘链接)
笔试面试题目:求缺失的最小正整数
What is the difference between vivoy73s and vivoy70s
That's what software testing is all about?!
一个方案提升Flutter内存利用率
Spotify是如何推动数据驱动决策的?
C language I blog assignment 03
解决Safari浏览器下载文件文件名称乱码的问题
Architect (November 2020)
Top 5 Chinese cloud manufacturers in 2018: Alibaba cloud, Tencent cloud, AWS, telecom, Unicom
Deeplight Technology Bluetooth protocol SRRC certification services
PCIe 枚举过程