Sampling allocation bytes precisely without compromising the performance #9745
Conversation
|
@amicic @dmitripivkine please review changes, thanks |
| _vmThread->nonZeroHeapTop = tlh->realHeapTop; | ||
| tlh->realHeapTop = NULL; | ||
| } | ||
| } |
There was a problem hiding this comment.
Looks like the question I asked about support for dual TLH mode applies here: Am I reading correctly that if we set size for both TLHs we can get 2 * size to be allocated potentially?
There was a problem hiding this comment.
yes, we might not get very accurate result on nonzero case,have not found the better way to handle the non zero case.
There was a problem hiding this comment.
Do we have a platform that can actually have both active (btw, is X using just nonZeroHeapAlloc/Top?)? I'm willing to ignore that issue for now, if we think it will take more than a day to do/test it and we can follow up after the upcoming release (by as you said disabling one of two TLHs).
There was a problem hiding this comment.
I believe non-Zeroed TLH is using on pLinux (LE and BE) and AIX. I am not sure about current status of zLinux.
There was a problem hiding this comment.
Non-zero TLH can be used for primitive arrays only. So to see a mismatch an application should include primitive arrays to the allocation mix.
There was a problem hiding this comment.
I agree that proper handling for dual TLH case can be done later in separate change
LinHu2016
force-pushed
the
JEP331_update
branch
9 times, most recently
from
3e18140
to
86eb958
Compare
in order to sampling heap allocation bytes precisely without compromising the performance, we have the below changes. Handle instrumentableAllocateHook and VM_OBJECT_ALLOCATE_WITHIN_THRESHOLD is still via disabling inline allocation Handle smapling for tracepoint is still during out of line allocation Handle smapling for JEP331 is via setTLHSamplingTop(size) Using fake Heap Top instead of fake Heap Alloc for disabling inline allocation (realHeapAlloc-->realHeapTop, set/getRealAlloc()-->set/getRealTop(), getRealSize(), getUsedSize()) Using fake Heap Top to force out of line allocation at sampling thresold for sampling heap allocation (setTLHSamplingTop()/resetTLHSamplingTop()) setTLHSamplingTop(size) are only called in the below 3 cases 1, sampling threshold has been changed via GC-VM api j9gc_set_allocation_sampling_interval() 2, TLH is refreshed 3, after sampling is done Counting trace allocation byte includes allocation bytes inside TLH Cache before flushing(_stats.bytesAllocated(true), stats->_tlhAllocatedUsed, ) Handle traceAllocationByte for Health Center(_oolTraceAllocationBytesForTracepoint, oolObjectSamplingBytesGranularityForTracepoint) and traceAllocationByte for JEP331(_traceAllocationBytesForHook, objectSamplingBytesGranularityForHook) independently Signed-off-by: Lin Hu <linhu@ca.ibm.com>
|
Jenkins test sanity all jdk11 |
|
has verified the latest personal build with customer's JEP331 test |
| U8Pointer realHeapAlloc = adjustedToRange(vmThread.allocateThreadLocalHeap().realHeapAlloc(), base, top); | ||
| if(realHeapAlloc.notNull() && isSomethingToAdd(realHeapAlloc, heapTop)) { | ||
| excludedRangeList.add(new U8Pointer[] {realHeapAlloc, heapTop}); | ||
| U8Pointer realHeapTop = adjustedToRange(vmThread.allocateThreadLocalHeap().realHeapTop(), base, top); |
There was a problem hiding this comment.
This will fail with NoSuchFieldError when examining core files created before the addition of realHeapTop.
There was a problem hiding this comment.
Good point, thank you very much.
in order to sampling heap allocation bytes precisely without
compromising the performance, we have the below changes.
Handle instrumentableAllocateHook and
VM_OBJECT_ALLOCATE_WITHIN_THRESHOLD is still via disabling inline
allocation
Handle smapling for tracepoint is still during out of line allocation
Handle smapling for JEP331 is via setTLHSamplingTop(size)
Using fake Heap Top instead of fake Heap Alloc for disabling inline
allocation (realHeapAlloc-->realHeapTop,
set/getRealAlloc()-->set/getRealTop(), getRealSize(), getUsedSize())
Using fake Heap Top to force out of line allocation at sampling thresold
for sampling heap allocation (setTLHSamplingTop()/resetTLHSamplingTop())
setTLHSamplingTop(size) are only called in the below 3 cases
1, sampling threshold has been changed via GC-VM api
j9gc_set_allocation_sampling_interval()
2, TLH is refreshed
3, after sampling is done
Counting trace allocation byte includes allocation bytes inside TLH
Cache before flushing(_stats.bytesAllocated(true),
stats->_tlhAllocatedUsed, )
Handle traceAllocationByte for Health
Center(_oolTraceAllocationBytesForTracepoint,
oolObjectSamplingBytesGranularityForTracepoint) and traceAllocationByte
for JEP331(_traceAllocationBytesForHook,
objectSamplingBytesGranularityForHook) independently
depend on eclipse-omr/omr#5260
fix: #7740
Signed-off-by: Lin Hu linhu@ca.ibm.com