RFR 8042668: GC Support for shared heap ranges in CDS (RE: JDK-8059092)

[Jiangli Zhou]

Hi Tom and Per,

On Jun 1, 2015, at 8:22 AM, Tom Benson <tom.benson at oracle.com> wrote:

> Hi Per,
> Thanks very much for the review.
> 
> On 6/1/2015 10:35 AM, Per Liden wrote:
>> Hi Tom,
>> 
>> On 2015-05-29 23:30, Tom Benson wrote:
>>> Hi,
>>> Please review these changes for JDK-8042668, which constitute the GC
>>> support for JDK-8059092 for storing interned strings in CDS archives
>>> (JEP 250).  The RFR for JDK-8059092 was recently posted by Jiangli Zhou,
>>> and it would be best if overall comments could go to that thread, with
>>> GC-specific comments here.
>>> 
>>> JBS:   https://bugs.openjdk.java.net/browse/JDK-8042668
>>> Webrev: http://cr.openjdk.java.net/~brutisso/8042668/webrev.00/
>> 
>> 
>> Maybe it's just me, but the concept of "recording" feels a bit strange in this context. May I suggest that we remove the "record" and "recording" part of the names and instead just call it an "archive" that we can allocate in? Something like:
>> 
>> class G1ArchiveAllocator ... {
>>  HeapWord* mem_allocate(...);
>> 
>>  void finalize(...);
>> 
>>  ...
>> };
>> 
>> class G1CollectedHeap ... {
>>  void begin_archive_mem_allocate();
>> 
>>  bool is_archive_mem_allocation_too_large(...);
>> 
>>  HeapWord* archive_mem_allocate(...);
>> 
>>  void end_archive_mem_allocate(...);
>> 
>>  ...
>> };
> 
> Hmmm....   Yes, I guess the name "RecordingAllocator" does show the evolution of the design, more than the ultimate use.  It was named "recording" because it allowed a way to keep track of the recorded ranges, in contrast with an earlier design that allocated a block of memory up front.   I'm fine with changing this to an ArchiveAllocator as you suggest, if I hear no objections.


Removing the “Recording” part sounds good to me too.

Thanks,
Jiangli

> 
>> 
>> 
>> g1CollectedHeap.cpp
>> -------------------
>> 
>> * In G1CollectedHeap::end_record_alloc_range(), shouldn't we delete the allocator as the last step?
>> 
> 
> Yes.  I think I made that change at one point and then removed it for some reason, which may be gone.  I'll re-make it.
> 
> 
>> * I guess this change could be skipped, as it makes the comment slightly malformed.
>> 
>> -        // We ignore humongous regions, we left the humongous set unchanged
>> +        // We ignore humongous regions.
>> +        // We left the humongous set unchanged,
>> 
> 
> OK.
> 
>> 
>> g1Allocator.hpp
>> ---------------
>> 
>> + class G1RecordingAllocator : public CHeapObj<mtGC> {
>> +   friend class VMStructs;
>> 
>> You could skip this friend declaration, since it's not accessed by VMStructs. Only needed if the class is exposed in the SA.
>> 
>> 
> 
> OK.  Thanks,
> Tom
> 
>> cheers,
>> /Per
>> 
>>> 
>>> These changes add a new "archive" region type to G1.  The description
>>> field in JDK-8042668 contains an "Implementation Notes" section which
>>> describes components of the design, and should be useful for a code
>>> review.   The overview:
>>> 
>>>    "Archive" regions are G1 regions that are not modifiable by GC,
>>>    being neither scavenged nor compacted, or even marked in the object
>>>    header. They can contain no pointers to non-archive heap regions,
>>>    and object headers point to shared CDS metaspace (though this last
>>>    point is not enforced by G1). Thus, they allow the underlying
>>>    hardware pages to be shared among multiple JVM instances.
>>> 
>>>    In short, a dump-time run (using -Xshare:dump) will allocate space
>>>    in the Java heap for the strings which are to be shared, copy the
>>>    string objects and arrays to that space, and then archive the entire
>>>    address range in the CDS archive. At restore-time (using
>>>    -Xshare:on), that same heap range will be allocated at JVM init
>>>    time, and the archived data will be mmap'ed into it. GC must treat
>>>    the range as 'pinned,' never moving or writing to any objects within
>>>    it, so that cross-JVM sharing will work.
>>> 
>>> Testing:  All testing for JDK-8059092 included this code. Manual
>>> testing with prototype calls to the new GC support was performed before
>>> integration, along with JPRT and benchmark runs.
>>> 
>>> Performance:  The GC changes had no significant impact on SpecJBB, JVM,
>>> or Dacapo benchmarks, run on x64 Linux.  However, a small (~1%) increase
>>> in Full GC times was seen in tests when the shared string support was
>>> not in use, when runs are configured to encounter them.   When shared
>>> strings ARE in use, the impact could be as high as 5% for a likely
>>> worst-case.   Please see the JBS entry for a discussion.
>>> 
>>> Thanks,
>>> Tom
>>> 
>