Reputation: 893
I've been playing around with the Aeson parser's benchmark suite and got some surprising results comparing their strict parser and the lazy one:
AesonParse
benchmark on citylots.json
takes dramatically more time and space than other datasets of comparable size. To get the benchmark running:
cd aeson; cabal install
cd benchmark; make
./AesonParse_p [blksz] [runs] [path-to-json] +RTS -p -hc
Here are some profiles I got with +RTS
turned on:
From the profiles the leak traced back to jstring_
in Data.Aeson.Parser.Internal
Anyone knows what's going on?
Sun Nov 15 13:56 2015 Time and Allocation Profiling Report (Final)
AesonParse_p +RTS -p -hc -RTS 105532 700 json-data/jp100.json
total time = 3.42 secs (3417 ticks @ 1000 us, 1 processor)
total alloc = 4,191,614,072 bytes (excludes profiling overheads)
COST CENTRE MODULE %time %alloc
jstringbang_ Data.Aeson.Parser.Internal 96.4 93.1
arrayValues Data.Aeson.Parser.Internal 1.7 1.9
object_' Data.Aeson.Parser.Internal 1.1 1.9
main.\.\.loop.refill Main 0.7 3.2
individual inherited
COST CENTRE MODULE no. entries %time %alloc %time %alloc
MAIN MAIN 116 0 0.0 0.0 100.0 100.0
main Main 233 0 0.0 0.0 100.0 100.0
main.blkSize Main 242 1 0.0 0.0 0.0 0.0
main.count Main 239 1 0.0 0.0 0.0 0.0
main.\ Main 234 1 0.0 0.0 100.0 100.0
main.\.\ Main 235 1 0.0 0.0 100.0 100.0
main.\.\.rate Main 252 1 0.0 0.0 0.0 0.0
main.\.\.loop Main 238 701 0.1 0.0 100.0 100.0
object_' Data.Aeson.Parser.Internal 245 0 1.1 1.9 99.2 96.8
jstring_ Data.Aeson.Parser.Internal 247 0 96.4 93.1 98.1 94.9
array_' Data.Aeson.Parser.Internal 249 0 0.0 0.0 1.7 1.9
arrayValues Data.Aeson.Parser.Internal 251 0 1.7 1.9 1.7 1.9
main.\.\.loop.refill Main 241 1 0.7 3.2 0.7 3.2
CAF GHC.IO.FD 217 0 0.0 0.0 0.0 0.0
CAF GHC.IO.Encoding.Iconv 215 0 0.0 0.0 0.0 0.0
CAF Text.Read.Lex 209 0 0.0 0.0 0.0 0.0
CAF GHC.IO.Handle.FD 202 0 0.0 0.0 0.0 0.0
CAF GHC.Conc.Signal 198 0 0.0 0.0 0.0 0.0
CAF GHC.IO.Handle.Text 191 0 0.0 0.0 0.0 0.0
CAF GHC.IO.Encoding 188 0 0.0 0.0 0.0 0.0
CAF Data.Time.Clock.UTC 129 0 0.0 0.0 0.0 0.0
CAF Data.Aeson.Parser.Internal 124 0 0.0 0.0 0.0 0.0
array_' Data.Aeson.Parser.Internal 248 1 0.0 0.0 0.0 0.0
arrayValues Data.Aeson.Parser.Internal 250 1 0.0 0.0 0.0 0.0
jstringbang_ Data.Aeson.Parser.Internal 246 1 0.0 0.0 0.0 0.0
object_' Data.Aeson.Parser.Internal 244 1 0.0 0.0 0.0 0.0
CAF Main 123 0 0.0 0.0 0.0 0.0
main Main 232 1 0.0 0.0 0.0 0.0
main.blkSize Main 243 0 0.0 0.0 0.0 0.0
main.count Main 240 0 0.0 0.0 0.0 0.0
main.\ Main 236 0 0.0 0.0 0.0 0.0
main.\.\ Main 237 0 0.0 0.0 0.0 0.0
Upvotes: 1
Views: 153
Reputation: 52049
From what I can tell this is should be expected.
The file citylots.json is a 167 MB file consisting of a single Object. The AesonParse program is building the entire object in memory, and that explains the memory ramp in the profile.
By contrast, the files companies.json and enron.json at http://jsonstudio.com/resources/ are "line-oriented" JSON files - each line is a JSON object and there are no commas between the objects. When you run AesonParse on either of these files it is only reading the first line.
Upvotes: 1