[
https://issues.apache.org/jira/browse/HBASE-17747?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15923660#comment-15923660
]
Yu Li commented on HBASE-17747:
-------------------------------
bq. Again, could you please share your GC options?
Sure, nothing special actually...
{noformat}-XX:+UseConcMarkSweepGC -Xmx40g -Xms40g -Xmn8g
-XX:SurvivorRatio=2{noformat}
And we tried to tune Xmn and SurvivorRation but no combination worked. Just try
the embedded mode out, I bet you'll see the issue I describe here that cannot
be fixed through tuning GC options.
bq. Weak reference is also suitable here in theory...
But only in theory, and in real world the dream breaks... And with soft
reference we have 5%-7% perf bump up even in distributed mode...
Let me list the commits I tried to make it live with weak reference (but
failed) besides tuning GC options:
{noformat}
commit 98a6c742ed4f5a8307f40df2bd95504ce1eb0444
Author: Yu Li <[email protected]>
Date: Thu Dec 29 17:06:59 2016 +0800
Try to accelerate weak reference purge for WeakObjectPool
Introduced a blocking queue, one producer to read from ReferenceQueue and
enqueue, and multiple consumers
read from blocking queue then remove from ConcurrentHashMap in parallel
commit 0e237d345b9c6b9fbc9edd597b727139b172a15a
Author: Yu Li <[email protected]>
Date: Thu Dec 29 15:08:28 2016 +0800
Add support of using exclusive thread for reference purge in WeakObjectPool
and use it in IdReadWriteLock (addendum)
commit 22add08c423f07058a2e231cec87806878c0d513
Author: Yu Li <[email protected]>
Date: Thu Dec 29 14:49:24 2016 +0800
Add support of using exclusive thread for reference purge in WeakObjectPool
and use it in IdReadWriteLock
commit a9e1c9bef1b94e56d6ddda5fcbe7d7c52f5e9da4
Author: Yu Li <[email protected]>
Date: Thu Dec 29 11:22:39 2016 +0800
Reduce invocation rate of WeakObjectPool#purge in IdReadWriteLock#getLock
to avoid contention
{noformat}
So I did tried before using soft reference...
> Support both weak and soft object pool
> --------------------------------------
>
> Key: HBASE-17747
> URL: https://issues.apache.org/jira/browse/HBASE-17747
> Project: HBase
> Issue Type: Improvement
> Affects Versions: 2.0
> Reporter: Yu Li
> Assignee: Yu Li
> Fix For: 2.0
>
> Attachments: HBASE-17747.patch, HBASE-17747.v2.patch,
> HBASE-17747.v3.patch
>
>
> During YCSB testing on embedded mode after HBASE-17744, we found that under
> high read load GC is quite severe even with offheap L2 cache. After some
> investigation, we found it's caused by using weak reference in
> {{IdReadWriteLock}}. In embedded mode the read is so quick that the lock
> might already get promoted to the old generation when the weak reference is
> cleared, which causes dirty card table (old reference get removed and new
> lock object set into {{referenceCache}}, see {{WeakObjectPool#get}}) thus
> slowing YGC. In distributed mode there'll also be more lock object created
> with weak reference than soft reference that slowing down the processing.
> So we proposed to use soft reference for this {{IdReadWriteLock}} used in
> cache, which won't get cleared until JVM memory is not enough, and could
> resolve the issue mentioned above. What's more, we propose to extend the
> {{WeakObjectPool}} to be more generate to support both weak and soft
> reference.
> Note that the GC issue only emerges under embedded mode with DirectOperator,
> in which case all costs on the wire is removed thus produces extremely high
> concurrency.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
