I have updated the Travis build file for my application to use Racket-on-Chez as one of the build configurations and, while the code builds and automated tests pass, the performance is very poor. I know that measuring times of Travis builds is not exactly scientific, however, I have run these builds four times and there is not much variations in the numbers. Here is a summary of my findings (Racket CS is the build from Dec 6, 2018):
* Racket CS runs out of memory when assembling a distribution (a "raco dist" command) -- my current build script runs the equivalent of "raco make", "raco exe" and "raco dist" from the same racket process, perhaps memory is not released? * Build time for my application, "raco make", is 98 seconds for Racket 7.1 and 187 seconds for Racket CS * `df-test`, a test that runs mostly Racket code, has increased from approximatly 5 seconds in Racket 7.1 to about 14 seconds in Racket CS. * `db-upgrade-test` is a test that mostly runs SQL code inside SQLite, the racket code just reads SQL statements and executes them. This test runs in the same amount of time. * the remaining tests, `db-test`, `trends-test` `aggregate-test` and `fit-test` all have running times that are about 50% larger in Racket CS when compared to Racket 7.1. These test spend a significant amout of time running SQL code which has not changed in timing, so the performance drop of the Racket code is probably worse than that. None of the above are actual performance tests -- they are meant to verify that the application works correctly, they just happen to run for a long time even in Racket 7.1 and they make perfromance problems show up more clearly. I have summarized my results here: https://docs.google.com/spreadsheets/d/1r1ZKtw4lDfv5xGXCfrOuE6jlCAMm0atMic4BfAnLLLw/edit?usp=sharing You can also find the build logs below, Travis adds timings for each step and the builds are 209, 210, 211 -- the first round in the spread sheet is a restart of 209 and unfortunately Travis has not kept the logs for the original build: https://travis-ci.org/alex-hhh/ActivityLog2/builds Is Racket-on-Chez built in debug mode? The installation size for Racket CS is 1.9 Gb while the installation size for Racket 7.1 is approximately 0.5 Gb... I have not installed and run Racket CS on my machine, so I don't know if the final application runs and what the user level performance is. Best Regards, Alex. On Tuesday, November 20, 2018 at 6:06:17 AM UTC+8, Matthew Flatt wrote: > > The Utah snapshot site now includes Racket-on-Chez builds for x86_64 > Windows, Mac OS, and Linux: > > http://pre.racket-lang.org/ > -> follow the "University of Utah" link > -> scroll down to "Racket CS" > > Although Racket-on-Chez is not yet ready to replace the existing > implementation of Racket for most purposes, it should generally work > now --- and it might even be useful for some purposes. Many, many > details have to be right for Racket-on-Chez to assemble itself into a > snapshot distribution, so this is a significant milestone. > > Try it out. Report bugs at > > https://github.com/racket/racket/issues > > > Here are some issues that we already know about: > > * Load times can be slower. > > * Compile times can be slower, including package installation. > > * Errors that originate from Chez Scheme have non-Rackety messages. > Values in those error message can look especially strange. > > * Not all `racket` command-line arguments are handled, yet. > > * Not all core Racket tests pass, yet. > > * No cross-compilation, yet. > > * No single-precision flonums, no extflonums. > > -- You received this message because you are subscribed to the Google Groups "Racket Users" group. To unsubscribe from this group and stop receiving emails from it, send an email to racket-users+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
