meiersi

import qualified Data.ByteString      as S
import qualified Data.ByteString.Lazy as L

import Blaze.ByteString.Builder
import Blaze.ByteString.Builder.Char.Utf8 (fromString)

import Data.Monoid

infixl 4 <>

meiersi

running benchmarks for 1 chunk size in bytes
collecting 100 samples, 1 iterations each, in estimated 24.22919 s
collecting 100 samples, 1 iterations each, in estimated 4.534197 s
collecting 100 samples, 1 iterations each, in estimated 5.621886 s

ranking for 1 chunk size in bytes
  1. compaction only: mean 3.469537 ms (2p 3.352643 ms, 98p 4.173089 ms, out 10)
  2. with compaction: mean 6.508811 ms (2p 6.260396 ms, 98p 7.141438 ms, out  4)
  3. direct:          mean 138.4184 ms (2p 128.6385 ms, 98p 154.6016 ms, out  0)

meiersi

┌─(simon@bumped)-(~/repositories/git/github/meiersi/aeson/benchmarks)-(master ?)
└─╼ ./AesonEncodeOld 500 json-data/*.json                                                          
json-data/example.json:
  0.047853s
  10449 per second
json-data/integers.json:
  0.123015s
  4065 per second
json-data/jp100.json:
  0.937498s

meiersi

-- Required for the MaybeC example below.
{-# LANGUAGE Rank2Types #-}

-- | HaskellerZ sketch-book from Juli 11th, 2011
module HaskellerZ where

import qualified Data.Map as M
import Control.Monad.State
import Control.Monad.Identity()

meiersi

benchmarking decode/en/aeson
mean: 1.549123 ms, lb 1.519203 ms, ub 1.589703 ms, ci 0.950
std dev: 177.1128 us, lb 137.1106 us, ub 225.7957 us, ci 0.950
found 10 outliers among 100 samples (10.0%)
  3 (3.0%) high mild
  7 (7.0%) high severe
variance introduced by outliers: 83.157%
variance is severely inflated by outliers

meiersi

./dist/build/bench-bytestring-builder/bench-bytestring-builder snoyman          
Sanity checks:
 lengths of input data: [10001,10001,10001,10001,10001,10001,10001]

warming up
estimating clock resolution...
mean is 5.480878 us (160001 iterations)
found 3843 outliers among 159999 samples (2.4%)
  3403 (2.1%) high severe
estimating cost of a clock call...

meiersi

└─╼ ./cabal-dev/bin/cereal-bench                                                    
Sanity checks:
 lengths of input data: [11,11,11,11,11,11,11]

cereal      :/[Int]: 96
binary      :/[Int]: 96
new builder :/[Int]: 100
new builder2:/[Int]: 96
cereal      :/[String]: 138
binary      :/[String]: 138

meiersi

-- |
-- Copyright   : (c) 2012 Simon Meier
-- License     : BSD3-style (see LICENSE)
--
-- Maintainer  : Simon Meier <iridcode@gmail.com>
-- Stability   : experimental
-- Portability : tested on GHC only
--
-- Benchmark decimal encoding speed of 10 million integers.
--

meiersi

-- Copyright   : (c) 2012 Michael Snoyman, Simon Meier
-- License     : BSD3-style (see LICENSE)
--
-- Maintainer  : Simon Meier <iridcode@gmail.com>
-- Stability   : experimental
-- Portability : tested on GHC only
--
-- Benchmark hexadecimal encoding speed. Use the preview repository of the new
-- bytestring builder to test this code:
--

meiersi

benchmarked against:  https://github.com/meiersi/blaze-binary/commit/d2bc3f004f608ebdd1eece624f288a263753bb36
and https://github.com/kolmodin/binary/commit/abbf46373620b45caca64e1072b300379e9c0fa1

warming up
estimating clock resolution...
mean is 1.466016 us (640001 iterations)
found 3590 outliers among 639999 samples (0.6%)
  2943 (0.5%) high severe
estimating cost of a clock call...
mean is 36.12254 ns (13 iterations)