ASCII mode file reading microbenchmark
1 view
Skip to first unread message
Jon
Nov 1, 2011, 10:35:46 AM11/1/11
to theco...@googlegroups.com
Some very early results (non RAMdisk based) from using the measurements repo on my Win7 32bit and Arch Linux (on VirtualBox). I'm reviewing both the benchmark runner and microbenchmark workload code to convince myself they're valid. Please do the same.
These results are from reading a 500,000 line, LF delimited file three times using `File.open('...', 'r')`
Here's the result from running Arch on my Win7 32bit via VirtualBox. IIRC, on *nix the reads will be performed in binary mode regardless of `File#open` setting. If true, this Linux result should probably be compared with Windows via `rci bench core_brd_filelines_lf`. However, since a user would not typically modify (a la C #ifdef) their Ruby to use 'r' when running on Linux and 'rb' when running on Windows, this comparison is still valid.
[jon@archee measurements-git]$ ./rci bench core_rd_filelines_lf
ruby 2.0.0dev (2011-10-31 trunk 33590) [i686-linux]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 0.880000 0.030000 0.910000 ( 0.908138)
----------------------------------------------- total: 0.910000sec
user system total real
core_rd_filelines_lf 0.930000 0.030000 0.960000 ( 0.959682)
For reference, here's the binary read performance on my 1.9.3-p0; about the same as Linux:
C:\projects\measurements-git>rci bench core_brd_filelines_lf
ruby 1.9.3p0 (2011-10-30 revision 33569) [i386-mingw32]
Rehearsal ---------------------------------------------------------
core_brd_filelines_lf 0.780000 0.171000 0.951000 ( 1.036060)
------------------------------------------------ total: 0.951000sec
user system total real
core_brd_filelines_lf 0.843000 0.156000 0.999000 ( 1.003057)
And here are the results from running different Rubies on my Win7 32bit:
C:\projects\measurements-git>pik run rci bench core_rd_filelines_lf
jruby 1.6.5 (ruby-1.9.2-p136) (2011-10-25 9dcd388) (Java HotSpot(TM) Client VM 1.7.0_01) [Windows 7-x86-java]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 8.600000 0.000000 8.600000 ( 8.576000)
----------------------------------------------- total: 8.600000sec
user system total real
core_rd_filelines_lf 8.512000 0.000000 8.512000 ( 8.512000)
ruby 1.8.7 (2011-06-30 patchlevel 352) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 1.826000 0.218000 2.044000 ( 2.084764)
----------------------------------------------- total: 2.044000sec
user system total real
core_rd_filelines_lf 1.888000 0.172000 2.060000 ( 2.068263)
ruby 1.9.2p290 (2011-07-09) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 18.268000 0.062000 18.330000 ( 18.333328)
---------------------------------------------- total: 18.330000sec
user system total real
core_rd_filelines_lf 18.034000 0.031000 18.065000 ( 18.084296)
ruby 1.9.3dev (2011-10-11 revision 33457) [i386-mingw32]
ruby 1.9.3p0 (2011-10-30 revision 33569) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 16.863000 0.062000 16.925000 ( 16.932150)
---------------------------------------------- total: 16.925000sec
user system total real
core_rd_filelines_lf 16.677000 0.063000 16.740000 ( 16.780130)
ruby 2.0.0dev (2011-10-31 trunk 33590) [i386-mingw32]
Rehearsal --------------------------------------------------------
core_rd_filelines_lf 16.972000 0.094000 17.066000 ( 17.094170)
---------------------------------------------- total: 17.066000sec
user system total real
core_rd_filelines_lf 16.973000 0.047000 17.020000 ( 17.045164)
Jon
---
twitter: @jonforums
rogerdpack
Nov 1, 2011, 2:34:13 PM11/1/11
to TheCodeShop
Results are a bit mangled maybe you can put it into a gist?
I had once thought it would be nice to overwrite the default MRI I/O
with some hand-brewed windows friendlier stuff... :P
I had once thought it would be nice to overwrite the default MRI I/O
with some hand-brewed windows friendlier stuff... :P
Jon
Nov 1, 2011, 2:42:45 PM11/1/11
to TheCodeShop
On Nov 1, 2:34 pm, rogerdpack <rogerdp...@gmail.com> wrote:
> Results are a bit mangled maybe you can put it into a gist?
...switching the web view to use fixed font.
> Results are a bit mangled maybe you can put it into a gist?
> I had once thought it would be nice to overwrite the default MRI I/O
> with some hand-brewed windows friendlier stuff... :P
Jon
Nov 1, 2011, 2:59:33 PM11/1/11
to TheCodeShop
> > Results are a bit mangled maybe you can put it into a gist?
>
> ...switching the web view to use fixed font.
>
my IE9 (surprisingly) doesn't pay attention to Google's settings....
>
> ...switching the web view to use fixed font.
>
http://pastie.org/2794875
Roger Pack
Nov 1, 2011, 3:18:35 PM11/1/11
to theco...@googlegroups.com
Also interesting is (I think) you get the same poor slowdown when you specify in Linux
File.open(filename, 'ra') or the like.
-r
File.open(filename, 'ra') or the like.
-r
Reply all
Reply to author
Forward
0 new messages