Re: [ANN] lua-marshal - fast table serialization
[
Date Prev][
Date Next][
Thread Prev][
Thread Next]
[
Date Index]
[
Thread Index]
- Subject: Re: [ANN] lua-marshal - fast table serialization
- From: Richard Hundt <richardhundt@...>
- Date: 2011年1月03日 22:01:34 +0100
On 01/03/2011 02:48 PM, Alexander Gladysh wrote:
I've updated my silly serialization benchmark (attached). Lua-Marshal is
fast enough to be considered (by me) a viable alternative to my luabins
for cases when I'd need to serialize cyclic tables and functions. (I
never really needed to serialize them though.)
BTW, I remember that Fabien wanted to serialize such stuff for Metalua.
Thanks for these. I added pluto to the mix and here's the result:
-------------------------------------------------------------------
name | rel | abs s / iter = us (1e-6 s) / iter
-------------------------------------------------------------------
luabins_load | 1.0000 | 0.42 / 100000 = 4.200000 us
luabins_save | 1.0952 | 0.46 / 100000 = 4.600000 us
marshal | 1.1429 | 0.48 / 100000 = 4.800000 us
unengram | 1.4524 | 0.61 / 100000 = 6.100000 us
unmarshal | 1.7143 | 0.72 / 100000 = 7.200000 us
loadstring | 3.7857 | 1.59 / 100000 = 15.900000 us
pluto_unpersist | 3.9762 | 1.67 / 100000 = 16.700000 us
pluto_persist | 5.8571 | 2.46 / 100000 = 24.600000 us
concat | 9.3810 | 3.94 / 100000 = 39.400000 us
engram | 42.1190 | 17.69 / 100000 = 176.900000 us
So I guess from luabins through marshal to pluto we've got a reasonable
spectrum from light-weight to heavy duty, with proportional performance.
Engram is too slow to be considered even marginally useful.
Serialization to Lua source is much faster (and more secure).
This engram implementation remains a toy until I (or somebody) rewrites
it in C, but even then, it probably has limited value, considering the
alternatives.
I thought I'd share it in case somebody wants to have some fun with
generating standard Lua 5.1 bytecode and they want some code to start
with (hence my comment "Anyway, just a bit of fun." in my original
mail). I heartily recommend KHMan's++ "A No-Frills Introduction to Lua
5.1 VM Instructions" if this sort of thing interests you [1]
Cheers,
Richard
[1]
http://luaforge.net/docman/view.php/83/98/ANoFrillsIntroToLua51VMInstructions.pdf
git clone https://github.com/agladysh/luamarca
$ ./run_benchmark.sh benchmark.lua 1e5
lua
-------------------------------------------------------------------
name | rel | abs s / iter = us (1e-6 s) / iter
-------------------------------------------------------------------
luabins_load | 1.0000 | 0.57 / 100000 = 5.700000 us
luabins_save | 1.1404 | 0.65 / 100000 = 6.500000 us
marshal | 1.1754 | 0.67 / 100000 = 6.700000 us
unmarshal | 1.7193 | 0.98 / 100000 = 9.800000 us
unengram | 2.0000 | 1.14 / 100000 = 11.400000 us
loadstring | 3.2807 | 1.87 / 100000 = 18.700000 us
concat | 9.1754 | 5.23 / 100000 = 52.300000 us
tstr | 12.2982 | 7.01 / 100000 = 70.100000 us
tserialize | 17.0351 | 9.71 / 100000 = 97.100000 us
engram | 38.3860 | 21.88 / 100000 = 218.800000 us
The full benchmark below was run with bench.engram/unengram commented out.
$ ./run_benchmark.sh benchmark.lua 1e6
lua
-------------------------------------------------------------------
name | rel | abs s / iter = us (1e-6 s) / iter
-------------------------------------------------------------------
luabins_load | 1.0000 | 5.68 / 1000000 = 5.680000 us
luabins_save | 1.1479 | 6.52 / 1000000 = 6.520000 us
marshal | 1.1954 | 6.79 / 1000000 = 6.790000 us
unmarshal | 1.5581 | 8.85 / 1000000 = 8.850000 us
loadstring | 3.3151 | 18.83 / 1000000 = 18.830000 us
concat | 9.4401 | 53.62 / 1000000 = 53.620000 us
tstr | 12.3644 | 70.23 / 1000000 = 70.230000 us
tserialize | 16.6021 | 94.30 / 1000000 = 94.300000 us
Modern LJ2 seems to be too smart for my benchmark (or I did break
something), so no results for LJ2 this time.
Alexander.
P.S. Disclaimer: read benchmark code, run it on your own box, write your
own benchmark — this one is a toy. But if you'd see a way to improve
something, please tell me.