Re: [15] RFR 8238633: JVMTI heap walk should consult GC for marking oops

[Zhengyu Gu]

Hi Serguei,

On 2/18/20 3:53 PM, serguei.spit...@oracle.com wrote:
Hi Zhengyu,

It looks okay to me.
The testing you do looks enough for verification.
But I'm not sure about performance testing though.

Thanks for reviewing.

I asked around my colleagues, if they knew any benchmarks for JVMTI heap 
walk, the answer was 'no'. As Stefan mentioned, this is a slow piece of 
code, I doubt if there is any benchmarks for it.

I would appreciate it if any performance people can chip in.

Thanks,

-Zhengyu





Thanks,
Serguei


On 2/17/20 6:51 AM, Zhengyu Gu wrote:
Hi Stefan,

Thanks for the review and suggestions, updated accordingly:

http://cr.openjdk.java.net/~zgu/JDK-8238633/webrev.01/


---
Previously, the calls to 'mark' and 'visited' were inlineable, but 
now every GC has to take a virtual call when marking the objects. My 
guess is that this code is slow anyway, and that it doesn't matter 
too much, but did you measure the effect of that change with, for 
example, G1?

I did rough measurement, timing 
vmTestbase/nsk/jvmti/unit/heap/HeapWalkTests/TestDescription.java test.

If you know any tests/benchmarks I should measure, please let me know.

Thanks,

-Zhengyu


Thanks,
StefanK

Test:
   hotspot_gc
   vmTestbase_nsk_jdi
   vmTestbase_nsk_jvmti

Thanks,

-Zhengyu