Re: [Jikesrvm-regression] The health of regression tests

[David P G. <gr...@us...>]

The machine is dedicated to Jikes RVM regressions, but it isn't in single 
user mode. (they do run in the middle of the night here, and those people 
who can log into the machine all know to stay off during the regression 
window).

We don't have a regular automated generation of compiler DNA because we 
have observed in the past that small changes in DNA don't actually cause 
the adaptive system to do anything that different and that the DNA tends 
to be somewhat stable (really a statement about the current lack of major 
development in the opt compiler).  There are scripts for generating DNA, 
so doing so on a regular basis would be easy, it's just not clear that it 
would be that valuable.  Could be useful for tracking regressions in 
compiler effectiveness/compile time, so it might be worth it from that 
perspective.  Impact on the adaptive system is likely to be fairly low. 
Tweaking DNA values by around 2X in either direction doesn't cause 
significant changes in overall system performance. 

--dave

> I've not been able to follow what's happening to the optimizing compiler 

> performance using the regression messages. If the machines the 
> regressions are being run on are heavily loaded isn't that going to skew 

> the meaning of the results too? Myself, I always benchmark in single 
> user mode to avoid unnecessary activities skewing benchmark results, but 

> even then the time differences between different runs can be 100s of 
> milliseconds (for what are runs of a few seconds). Related to this is 
> the status of the compiler DNA, has the generation of this been set up 
> to be updated as a regular job?
> 
> Regards,
> 
> Ian Rogers
> 
> 
> 
> -------------------------------------------------------
> SF.Net email is sponsored by: Discover Easy Linux Migration Strategies
> from IBM. Find simple to follow Roadmaps, straightforward articles,
> informative Webcasts and more! Get everything you need to get up to
> speed, fast. http://ads.osdn.com/?ad_id=7477&alloc_id=16492&op=click
> _______________________________________________
> Jikesrvm-regression mailing list
> Jik...@li...
> https://lists.sourceforge.net/lists/listinfo/jikesrvm-regression