Datamapper DataMapper → 1.1.1

Comments and changes to this ticket

• 

  Martin Gamsjaeger (snusnu) November 4th, 2010 @ 06:01 AM

  • State changed from “new” to “unconfirmed”
  • Tag set to “performance, regression”
• 

  Xavier Shay December 2nd, 2010 @ 11:26 PM

  As with #1442, a stand alone benchmarking script would be helpful.

• You flagged this item as spam.
  

  Alexander Sorokin January 5th, 2011 @ 05:20 PM

  Here's a standalone example. It's super-simple and gives almost 2x performance degradation with DM 1.0.2.

  https://gist.github.com/767201

  Thanks for the example template!

  Can you run the benchmark and confirm my timings?

• 

  Xavier Shay January 9th, 2011 @ 05:21 PM

  What version of dm-postgres-adapter are you using for 0.10? I can get all the other gems into a gemset using:
  

  gem install dm-core dm-migrations dm-timestamps --version "~> 0.10" --no-rdoc --no-ri
  

  My results for v1 showed a big difference between 1.8 and 1.9:

  == 1.0.2
  1.9: 1.600000   0.070000   1.670000 (  2.202385)
  1.8: 2.510000   0.110000   2.620000 (  3.133895)
  

• You flagged this item as spam.
  

  Alexander Sorokin January 9th, 2011 @ 09:20 PM

  We're using do_postgres-0.10.0 for 0.10 and ruby 1.8.

• You flagged this item as spam.
  

  Alexander Sorokin January 10th, 2011 @ 02:20 PM

  Yep. Ruby 1.9 is much faster. The gap with 0.10 is still there.

  == DM 0.10, ruby 1.9 0.560000 0.050000 0.610000 ( 1.470046)

• You flagged this item as spam.
  

  Alexander Sorokin January 10th, 2011 @ 02:21 PM

  Yep. Ruby 1.9 is much faster. The gap with 0.10 is still there.

  == DM 0.10, ruby 1.9

  0.560000 0.050000 0.610000 ( 1.470046)

• 

  Xavier Shay January 11th, 2011 @ 01:18 AM

  I can't seem to find that version of the gem anywhere. Can you attach it? I'll have a play around with some profiling tonight anyways.

• 

  Xavier Shay January 11th, 2011 @ 02:34 AM

  Saved a bit with a micro-optimization, but seems the biggest problems are structural? Would like to be wrong.
  Could be a win disabling assertions in production, they account for ~20% of the time.

• 

  Dan Kubb (dkubb) January 11th, 2011 @ 02:34 AM

  (from [733895177138b8674897deb6e806de468542b1ce]) Speed up #normalize_options by ~2x

  [#1443] https://github.com/datamapper/dm-core/commit/733895177138b8674897de...

• You flagged this item as spam.
  

  Alexander Sorokin January 11th, 2011 @ 12:29 PM

  I couldn't attach the gems here. I put them on S3: https://datamapper.s3.amazonaws.com/all_gems.tgz?AWSAccessKeyId=AKI...

• You flagged this item as spam.
  

  Alexander Sorokin January 12th, 2011 @ 06:06 PM

  Something has changed on my system, so that overall timings are better. There's an worrying trend however:

  DM 0.10.0 | 0.960000 0.070000 1.030000 ( 1.903432)
  DM 1.0.2 (gem fetch) | 2.480000 0.110000 2.590000 ( 3.622813)
  DM 1.0.2 (built from master) | 3.200000 0.110000 3.310000 ( 4.326419)

  (normalize_options makes it faster, but there seems to be something else that takes more time)

• You flagged this item as spam.
  

  Alexander Sorokin February 1st, 2011 @ 02:12 PM

  Do you guys have timeline for this? Do you plan to investigate this more?

• 

  Xavier Shay February 1st, 2011 @ 04:25 PM

  I plan to look at some general performance things such as iterating through collections, which I believe will be related to this. I don't have any time pressure though.

• You flagged this item as spam.
  

  Alexander Sorokin February 1st, 2011 @ 04:56 PM

  Thanks! We're blocked on these two tickets for our upgrade to DM 1.0.*. I hope that once these get resolved, we'll be able to switch over. We don't want to sit on 0.10 forever, that's why I'm checking on this :)

  Thanks a lot for the hard work! It's awesome!

• 

  Piotr Solnica (solnic) February 2nd, 2011 @ 03:35 AM

  I can also take a look and here are my results on 1.9.2p136: https://gist.github.com/807464 Seems like it's almost the same...

• 

  Dan Kubb (dkubb) March 16th, 2011 @ 11:07 PM

  • State changed from “unconfirmed” to “confirmed”
  • Milestone set to “1.1.1”
  • Milestone order changed from “196304” to “0”

  Alexander, even though we're about to release DatMapper 1.1 today I want you to know we haven't forgotten about this.

  I'm going to set one of our goals for 1.1.1 (or 1.2 if it requires minor api changes) to optimize the pathways for Model#new, Model#create and Model#all.

  In the meantime if you have any profiling output please feel free to attach it to this ticket. We'll be doing our own measuring, but it would be nice to have a baseline from you before we get started. (I saw the screenshots, but the full report would probably be most useful)