bench.rb

# frozen_string_literal: true

require 'benchmark/ips'
require 'active_support/all'
require 'zlib'

sizes = if ENV['SIZES']
  ENV['SIZES'].split(',').map(&:to_i)
else
  [0, 10, 2_000]
end

module ActiveSupport
  module Cache
    module Coders
      module OptimizedCoder
        extend self
        COMPRESSED = 0x01
        SERIALIZED = 0x02

        def dump(entry)
          dump_compressed(entry, false)
        end

        def dump_compressed(entry, threshold)
          flags = 0
          header = [entry.expires_at, entry.version]
          header.pop until header.empty? || !header.last.nil?
          header_payload = Marshal.dump(header)

          value = entry.value
          if value.is_a?(String)
            value_payload = entry.value
          else
            flags |= SERIALIZED
            value_payload = Marshal.dump(value)
          end

          if threshold && value_payload.bytesize >= threshold
            compressed_payload = Zlib::Deflate.deflate(value_payload)
            if compressed_payload.bytesize < value_payload.bytesize
              value_payload = compressed_payload
              flags |= COMPRESSED
            end
          end
          flags.chr << header_payload.bytesize.chr << header_payload << value_payload
        end

        def load(payload)
          flags, header_size = payload.unpack('CC')
          header = Marshal.load(payload.byteslice(2..-1))
          value_payload = payload.byteslice((2 + header_size)..-1)
          if flags & COMPRESSED > 0
            if flags & SERIALIZED > 0
              Entry.new(value_payload, compressed: true, expires_at: header[0], version: header[1])
            else
              Entry.new(Zlib::Inflate.inflate(value_payload), expires_at: header[0], version: header[1])
            end
          elsif flags & SERIALIZED > 0
            Entry.new(Marshal.load(value_payload), expires_at: header[0], version: header[1])
          else
            Entry.new(value_payload, expires_at: header[0], version: header[1])
          end
        end
      end
    end
  end
end

sizes.each do |bytesize|
  puts "================ #{bytesize} bytes ================"

  if bytesize == 0
    entry_marshal = Marshal.dump(ActiveSupport::Cache::Entry.new(''))
    entry_pack = ActiveSupport::Cache::Coders::Rails70Coder.dump(ActiveSupport::Cache::Entry.new('', compress: false))
    opt_entry_pack = ActiveSupport::Cache::Coders::OptimizedCoder.dump(ActiveSupport::Cache::Entry.new('', compress: false))
  else
    string = ('A'..'z').cycle.take(bytesize).join
    entry_marshal = Marshal.dump(ActiveSupport::Cache::Entry.new(string, expires_in: 1.hour, version: "v42").compressed(1_024))
    entry_pack = ActiveSupport::Cache::Coders::Rails70Coder.dump_compressed(ActiveSupport::Cache::Entry.new(string, expires_in: 1.hour, version: "v42", compress: false), 1_024)
    opt_entry_pack = ActiveSupport::Cache::Coders::OptimizedCoder.dump_compressed(ActiveSupport::Cache::Entry.new(string, expires_in: 1.hour, version: "v42", compress: false), 1_024)
  end

  puts "Marshal.dump.bytesize: #{entry_marshal.bytesize}"
  puts "EntryCoder.dump.bytesize: #{entry_pack.bytesize}"
  puts "OptimizedCoder.dump.bytesize: #{entry_pack.bytesize}"
  puts

  Benchmark.ips do |x|
    x.report('Marshal.dump') { Marshal.dump(ActiveSupport::Cache::Entry.new(string, expires_in: 1.hour, version: "v42").compressed(1_024)) }
    x.report('Rails70.dump') { ActiveSupport::Cache::Coders::Rails70Coder.dump_compressed(ActiveSupport::Cache::Entry.new(string, expires_in: 1.hour, version: "v42", compress: false), 1_024) }
    x.report('Optimized.dump') { ActiveSupport::Cache::Coders::OptimizedCoder.dump_compressed(ActiveSupport::Cache::Entry.new(string, expires_in: 1.hour, version: "v42", compress: false), 1_024) }
    x.compare!
  end

  puts
  Benchmark.ips do |x|
    x.report('Marshal.load') { Marshal.load(entry_marshal).value }
    x.report('Rails70.load') { ActiveSupport::Cache::Coders::Rails70Coder.load(entry_pack).value }
    x.report('OptimizedCoder.load') { ActiveSupport::Cache::Coders::OptimizedCoder.load(opt_entry_pack).value }
    x.compare!
  end
  puts
  puts
end

results.txt

================ 0 bytes ================
Marshal.dump.bytesize: 90
EntryCoder.dump.bytesize: 13
OptimizedCoder.dump.bytesize: 13

Warming up --------------------------------------
        Marshal.dump    17.129k i/100ms
        Rails70.dump    21.300k i/100ms
      Optimized.dump    16.390k i/100ms
Calculating -------------------------------------
        Marshal.dump    170.537k (± 1.4%) i/s -    856.450k in   5.023018s
        Rails70.dump    215.330k (± 1.3%) i/s -      1.086M in   5.045690s
      Optimized.dump    163.612k (± 1.4%) i/s -    819.500k in   5.009718s

Comparison:
        Rails70.dump:   215329.6 i/s
        Marshal.dump:   170537.1 i/s - 1.26x  (± 0.00) slower
      Optimized.dump:   163612.0 i/s - 1.32x  (± 0.00) slower


Warming up --------------------------------------
        Marshal.load    36.668k i/100ms
        Rails70.load    55.709k i/100ms
 OptimizedCoder.load    51.832k i/100ms
Calculating -------------------------------------
        Marshal.load    366.149k (± 1.4%) i/s -      1.833M in   5.008249s
        Rails70.load    550.303k (± 1.5%) i/s -      2.785M in   5.062743s
 OptimizedCoder.load    517.866k (± 2.0%) i/s -      2.592M in   5.006465s

Comparison:
        Rails70.load:   550303.1 i/s
 OptimizedCoder.load:   517866.1 i/s - 1.06x  (± 0.00) slower
        Marshal.load:   366148.8 i/s - 1.50x  (± 0.00) slower



================ 10 bytes ================
Marshal.dump.bytesize: 127
EntryCoder.dump.bytesize: 53
OptimizedCoder.dump.bytesize: 53

Warming up --------------------------------------
        Marshal.dump    14.675k i/100ms
        Rails70.dump    20.229k i/100ms
      Optimized.dump    19.653k i/100ms
Calculating -------------------------------------
        Marshal.dump    146.104k (± 0.9%) i/s -    733.750k in   5.022520s
        Rails70.dump    203.623k (± 1.1%) i/s -      1.032M in   5.067272s
      Optimized.dump    198.870k (± 1.2%) i/s -      1.002M in   5.040741s

Comparison:
        Rails70.dump:   203623.3 i/s
      Optimized.dump:   198870.5 i/s - same-ish: difference falls within error
        Marshal.dump:   146104.4 i/s - 1.39x  (± 0.00) slower


Warming up --------------------------------------
        Marshal.load    29.581k i/100ms
        Rails70.load    43.199k i/100ms
 OptimizedCoder.load    40.047k i/100ms
Calculating -------------------------------------
        Marshal.load    296.891k (± 1.1%) i/s -      1.509M in   5.082104s
        Rails70.load    431.656k (± 1.2%) i/s -      2.160M in   5.004583s
 OptimizedCoder.load    404.179k (± 5.1%) i/s -      2.042M in   5.072898s

Comparison:
        Rails70.load:   431656.3 i/s
 OptimizedCoder.load:   404178.5 i/s - 1.07x  (± 0.00) slower
        Marshal.load:   296890.8 i/s - 1.45x  (± 0.00) slower



================ 2000 bytes ================
Marshal.dump.bytesize: 221
EntryCoder.dump.bytesize: 127
OptimizedCoder.dump.bytesize: 127

Warming up --------------------------------------
        Marshal.dump     3.953k i/100ms
        Rails70.dump     4.023k i/100ms
      Optimized.dump     4.778k i/100ms
Calculating -------------------------------------
        Marshal.dump     39.823k (± 1.1%) i/s -    201.603k in   5.063150s
        Rails70.dump     40.129k (± 1.0%) i/s -    201.150k in   5.013145s
      Optimized.dump     47.953k (± 1.2%) i/s -    243.678k in   5.082282s

Comparison:
      Optimized.dump:    47953.3 i/s
        Rails70.dump:    40128.8 i/s - 1.19x  (± 0.00) slower
        Marshal.dump:    39822.7 i/s - 1.20x  (± 0.00) slower


Warming up --------------------------------------
        Marshal.load    10.688k i/100ms
        Rails70.load    13.666k i/100ms
 OptimizedCoder.load    12.594k i/100ms
Calculating -------------------------------------
        Marshal.load    107.654k (± 1.3%) i/s -    545.088k in   5.064150s
        Rails70.load    133.274k (± 2.6%) i/s -    669.634k in   5.028035s
 OptimizedCoder.load    124.683k (± 1.7%) i/s -    629.700k in   5.051781s

Comparison:
        Rails70.load:   133273.6 i/s
 OptimizedCoder.load:   124683.4 i/s - 1.07x  (± 0.00) slower
        Marshal.load:   107654.1 i/s - 1.24x  (± 0.00) slower