review(XXS): 7159766: Tiered compilation causes performance regressions

[Igor Veresov]

I think it's could be mostly a GC problem :) Namely it could because the lack of scalability of the code cache traversal. But I can't say for sure, I didn't really have a chance to analyze the benchmark in question. But I imagine it can have a huge number of nmethods, >=45k.. 

igor


On Friday, April 6, 2012 at 6:42 PM, Srinivas Ramakrishna wrote:

> Hi Igor, were the anomalies only wrt start-up, or were they also in longer run performance numbers? The bug's not yet visible on bugs.sun.com (http://bugs.sun.com), but I guess I can get the info later when the bug shows up there.
> 
> thanks!
> -- ramki
> 
> On Fri, Apr 6, 2012 at 4:14 PM, Igor Veresov <igor.veresov at oracle.com (mailto:igor.veresov at oracle.com)> wrote:
> > It's been decided to turn tiered compilation off by default in 7u4 because of the performance anomalies with Oracle FMW. 
> > 
> > Webrev: http://cr.openjdk.java.net/~iveresov/7159766/webrev.00/ (http://cr.openjdk.java.net/%7Eiveresov/7159766/webrev.00/)
> > 
> > Thanks!
> > igor
> > 
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.openjdk.java.net/pipermail/hotspot-compiler-dev/attachments/20120406/ccdef9a9/attachment.html