-
Notifications
You must be signed in to change notification settings - Fork 15
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Mandrel slower than corresponding GraalVM CE #288
Comments
The issue is possibly related to oracle#3697 |
After some more runs it looks like the difference is closer to 270s vs220s which is still significant. And the issue is reproducible with Mandrel 21.0.0.1-Final as well. |
Note the difference in heap space. Shorter runs use more heap. |
After some more experimentation it looks like this is related to the fact that Mandrel is not using libgraal. The results are listed in the table bellow (3 runs of each configuration):
It's clear that when GraalVM CE is run with This behavior is reproducible with prior Mandrel versions, so maybe we should consider whether Mandrel should start using libgraal as well. |
Using libgraal in Mandrel for Java 11 seems a no-go as libgraal depends heavily on JVMCI changes that are not available in OpenJDK 11. Moving to Java 17 we could reconsider this. |
Yes, I've reproduced this. ~20% slower build times. |
On the flip side I've seen Graal VM CE 21.2 (via
Graal VM CE 21.2 with
Mandrel 21.2 (via
|
For reference.
|
Resident set size is sometimes lower than reported Heap capacity due to lazy commit of the kernel. |
This issue appears to be stale because it has been open 30 days with no activity. This issue will be closed in 7 days unless |
Correction. Current implementation of memory reporting uses |
Running similar experiments with 22.3 I see that the difference is now smaller (~6% when comparing the average execution time of 5 runs). On these 5 runs Mandrel appears to have slightly bigger Peak RSS as well (~200MB on average). I think we can close this issue at this point. |
The ~6% difference still stands for 22.3.2 Raw numbers from 10 runs of 5 configurations:
|
In order to do "fair" comparisons, there might be other flags which could have some effect. Using
|
Closing this as the difference is in the ballpark of 5% and we haven't had any complaints from users. |
Description
Mandrel 21.2.0.0-Final appears to be slower than GraalVM 21.2.0 (311s vs 229s)
How To Reproduce
Steps to reproduce the behavior:
Configuration:
Expected behavior
Compilation times should be similar.
Additional context
The output of the first command (Mandrel) is:
The output of the second command (GraalVM) is:
The text was updated successfully, but these errors were encountered: