Elixir比Ruby慢吗？

Question

请帮助我解决有关Elixir与Ruby性能的基准问题.

我尝试在两种语言中实现相同的阶乘,而Ruby显示出比Elixir更好的结果:

# ruby_factorial_with_iterator.rb
def factorial_with_iterator(n)
  res = 1
  (1..n).each{|time| res *= time}
  res
end

p "factorial_with_iterator(200000)"
p factorial_with_iterator(200000)

运行后:

$ time ruby ruby_factorial_with_iterator.rb
real  0m18.378s
user  0m17.348s
sys   0m0.844s

和两个Elixir例子:

# elixir_factorial_with_iterator.exs
defmodule FactorialWithIterator do
  def of(n) do
    Enum.reduce(1..n, 1, &*/2)
  end
end

IO.puts "Factorial of 200000: "
IO.puts FactorialWithIterator.of(200000)

运行后:

$ time elixir elixir_factorial_with_iterator.exs
real  1m1.735s
user  1m1.556s
sys   0m0.104s

另一个例子:

# elixir_factorial_with_recursion.exs
defmodule FactorialWithRecursion do
  def of(0), do: 1
  def of(n) when n > 0 do
    n * of(n - 1)
  end
end

IO.puts "Factorial of 200000: "
IO.puts FactorialWithRecursion.of(200000)

运行后:

$ time elixir elixir_factorial_with_recursion.exs
real  1m7.149s
user  1m6.248s
sys   0m0.092s

为什么会有这么大的差异:Elixir - 1m1s和Ruby--只需18s？或者如何在Elixir中编写正确的迭代代码？

PS环境:

• Ubuntu 16.04
• ruby 2.3.1p112(2016-04-26修订版54768)[x86_64-linux]
• Erlang/OTP 19 [erts-8.3] [source-d5c06c6] [64-bit] [smp:4:4] [async-threads:10] [hipe] [kernel-poll:false]
• Elixir 1.4.4

Answer 1

正如其中一条评论所述,您正在使用它time,它也会启动VM,并且在elixir的情况下,将代码编译为BEAM字节码.为避免计算所有这些,您应该使用语言本身的基准测试工具.

我很好奇,所以我尝试自己对这些功能进行基准测试.

我用了:

• bencheeelixir:https://github.com/PragTob/benchee
• benchmark-ips对于ruby:https://github.com/evanphx/benchmark-ips

红宝石:

require 'benchmark/ips'

def factorial_with_iterator(n)
  res = 1
  (1..n).each{|time| res *= time}
  res
end

Benchmark.ips do |x|
  x.config(time: 5, warmup: 2)
  x.report('factorial_with_iterator.rb') do
    factorial_with_iterator(200000)
  end
  x.compare!
end

药剂:

defmodule Factorial do
  def iter(n) do
    Enum.reduce(1..n, 1, &*/2)
  end

  def recur(0), do: 1

  def recur(n) when n > 0 do
    n * recur(n - 1)
  end
end

Benchee.run(%{
  "factorial_with_iter.ex" => fn -> Factorial.iter(200000) end,
  "factorial_with_recur.ex" => fn -> Factorial.recur(200000) end
})

我得到了这些结果:

红宝石:

Warming up --------------------------------------
factorial_with_iterator.rb
                         1.000  i/100ms
Calculating -------------------------------------
factorial_with_iterator.rb
                          0.033  (± 0.0%) i/s -      1.000  in  29.994713s

药剂:

Name                              ips        average  deviation         median         99th %
factorial_with_iter.ex         0.0395        25.29 s     ±0.00%        25.29 s        25.29 s
factorial_with_recur.ex        0.0368        27.17 s     ±0.00%        27.17 s        27.17 s

Comparison:
factorial_with_iter.ex         0.0395
factorial_with_recur.ex        0.0368 - 1.07x slower

因此,这些结果显示Elixir稍微快一些,使用Ruby的实现需要大约30秒,而Elixir需要大约25秒和大约27秒.

但是,使用"每秒迭代次数",对于花费时间超过一秒的功能可能有点"错误".所以我也尝试了更低的输入.我用了1_000而不是200_000,得到了这些结果:

红宝石:

Warming up --------------------------------------
factorial_with_iterator.rb
                       169.000  i/100ms
Calculating -------------------------------------
factorial_with_iterator.rb
                          1.750k (± 8.0%) i/s -      8.788k in   5.064619s

药剂:

Name                              ips        average  deviation         median         99th %
factorial_with_recur.ex        3.15 K      317.36 ?s    ±12.72%         306 ?s      481.87 ?s
factorial_with_iter.ex         3.02 K      331.13 ?s    ±16.83%         316 ?s         559 ?s

Comparison:
factorial_with_recur.ex        3.15 K
factorial_with_iter.ex         3.02 K - 1.04x slower

奇怪的是,这表明Elixir比Ruby快得多.对于这两种实现,Elixir每秒能够执行超过3k次迭代,其中Ruby每秒只能执行1.75k迭代.

使用:

• ruby 2.4.1p111(2017-03-22修订版58053)[x86_64-darwin16]
• 在Erlang/OTP 21上运行的Elixir 1.6.5(用OTP 19编译)[erts-10.0] [source] [64-bit] [smp:8:8] [ds:8:8:10] [async-threads: 1] [hipe] [dtrace]