preform GC modifications

[Ofir Gordon]

Thank you very much for the detailed answer.
I'll take your recomendation to try and use a newer version, it just that
I'm working on a specific simulator which might have difficulties with
newer jdk versions, but it worth a try.

If it is ok I would like another followup question, regarding the actual
method in which the marking phase is preformed (again for the serial gc) -
when I follow after the calls from GenCollectedHeap::collect(), I can see
that finaly it gets in a method
GenCollectedHeap::gen_process_strong_roots(), where I believe the actual
marking happens? then inside the method it goes to
SharedHeap::process_strong_roots, and in there, there are many oops_do()
calls of different classes.
So my question is, where exactly the marking loop happens for this case?
and how can it be specified for different types of gc (cause not all go the
same code paths...)

Again, thanks a lot,
Ofir

‫בתאריך יום ד׳, 15 באפר׳ 2020 ב-15:14 מאת ‪Thomas Schatzl‬‏ <‪
thomas.schatzl at oracle.com‬‏>:‬

> On 15.04.20 12:17, Ofir Gordon wrote:
> > Hello,
> >
> > I'm trying to investigate the performance of different garbage
> collectors,
> > as part of a project I'm working on,
> > and as part of the project I need to preform modifications in the
> existing
> > GCs (working with openjdk8 hotspot).
>
> I really recommend working with at least jdk14 or even tip if possible.
> We removed quite a bit of cruft in the last six or so years. Also, the
> majority of people are currently working on tip, so you might get even
> less feedback  than otherwise when using such an old version. While the
> general rough structure is the same, probably your audience might not
> have a checkout handy, needing to check out that version, read into it,
> double check, etc.
>
> Particularly Serial GC got much easier to understand due to many many
> cleanups and removal of CMS.
>
> > I tried to start with a simple task and figure out where in the code the
> > call for the GC happens? I couldn't figure this out.
>
> Implementations of CollectedHeap::collect() perform the collection.
> GenCollectedHeap::collect() is the entry point for serial gc (and CMS).
>
> >
> > Also, I'm currently want to try only to run simple serial Mark&Sweep GC -
> > can someone please help me understand:
> > 1. how do I specify it when running the java program? (if I specify
> > -XX:+UseSerialGC is that enough? what type of GC is running for this
> flag?)
>
> Yes. Serial GC.
>
> > 2. is there a way to produce logs in which I can see that it is actually
> > working (more info than just when a GC is activated)
>
> Also in later versions logging is probably much better (maybe not for
> Serial GC) :)
>
> -XX:+PrintGC and -XX:PrintGCDetails, potentially with -XX:+Verbose
>
> > 3. most important - which class/methods in the source code are relevant
> for
> > that simple collector (basically I'm trying to understand which method
> > handles the marking phase and where is it being called)? I need this to
> try
> > modify some code and see that it is actually working.
>
> Serial gc is kind of strewn all over the place in jdk8, it only got
> better later. Start with defNewGeneration.*pp for the young collector
> and (gen)markSweep.*pp for the full collector iirc.
>
> Checking out jdk14+ and look for similiar files you can find in the
> src/hotspot/share/gc/serial directory in jdk8u might help here.
>
> >
> > I know that's a lot of questions, if someone can give me a hint on how to
> > start I will be very grateful.
> >
>
> Hth,
>    Thomas
>

‫בתאריך יום ד׳, 15 באפר׳ 2020 ב-15:14 מאת ‪Thomas Schatzl‬‏ <‪
thomas.schatzl at oracle.com‬‏>:‬

> On 15.04.20 12:17, Ofir Gordon wrote:
> > Hello,
> >
> > I'm trying to investigate the performance of different garbage
> collectors,
> > as part of a project I'm working on,
> > and as part of the project I need to preform modifications in the
> existing
> > GCs (working with openjdk8 hotspot).
>
> I really recommend working with at least jdk14 or even tip if possible.
> We removed quite a bit of cruft in the last six or so years. Also, the
> majority of people are currently working on tip, so you might get even
> less feedback  than otherwise when using such an old version. While the
> general rough structure is the same, probably your audience might not
> have a checkout handy, needing to check out that version, read into it,
> double check, etc.
>
> Particularly Serial GC got much easier to understand due to many many
> cleanups and removal of CMS.
>
> > I tried to start with a simple task and figure out where in the code the
> > call for the GC happens? I couldn't figure this out.
>
> Implementations of CollectedHeap::collect() perform the collection.
> GenCollectedHeap::collect() is the entry point for serial gc (and CMS).
>
> >
> > Also, I'm currently want to try only to run simple serial Mark&Sweep GC -
> > can someone please help me understand:
> > 1. how do I specify it when running the java program? (if I specify
> > -XX:+UseSerialGC is that enough? what type of GC is running for this
> flag?)
>
> Yes. Serial GC.
>
> > 2. is there a way to produce logs in which I can see that it is actually
> > working (more info than just when a GC is activated)
>
> Also in later versions logging is probably much better (maybe not for
> Serial GC) :)
>
> -XX:+PrintGC and -XX:PrintGCDetails, potentially with -XX:+Verbose
>
> > 3. most important - which class/methods in the source code are relevant
> for
> > that simple collector (basically I'm trying to understand which method
> > handles the marking phase and where is it being called)? I need this to
> try
> > modify some code and see that it is actually working.
>
> Serial gc is kind of strewn all over the place in jdk8, it only got
> better later. Start with defNewGeneration.*pp for the young collector
> and (gen)markSweep.*pp for the full collector iirc.
>
> Checking out jdk14+ and look for similiar files you can find in the
> src/hotspot/share/gc/serial directory in jdk8u might help here.
>
> >
> > I know that's a lot of questions, if someone can give me a hint on how to
> > start I will be very grateful.
> >
>
> Hth,
>    Thomas
>
>