[
https://issues.apache.org/jira/browse/HBASE-20074?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16376526#comment-16376526
]
Eshcar Hillel commented on HBASE-20074:
---------------------------------------
Here is a summary of the throughput results of write-only workload of past
benchmarks I ran with alpha-1code and more recent benchmarks Gali ran with
beta-2 code
|| ||CAM 0.02||CAM 0.1||lift||
||alpha-1|115K|110K|4%|
||beta-2|47.8K|48.8K|-2%|
We still cannot explain the degradation in performance and so it is hard to
say what would ultimately be the optimal setting.
Currently it seems CCM is performing better with 0.1 due to the usage of MSLABs.
Not sure we need to fix different config keys. If CCM is entengaled with MSLAB
and CAM is used whenever MSLAB is not used then we can get away with a single
config.
> [FindBugs] Same code on both branches in
> CompactingMemStore#initMemStoreCompactor
> ---------------------------------------------------------------------------------
>
> Key: HBASE-20074
> URL: https://issues.apache.org/jira/browse/HBASE-20074
> Project: HBase
> Issue Type: Bug
> Components: findbugs
> Reporter: stack
> Assignee: Gali Sheffi
> Priority: Major
>
> [~galish] Our findbugs checking was disabled for a few weeks and we just
> turned it on again. It found a good one. Mind fixing it please? Meantime I've
> undone the if class to get a factor dependent on index type. Thanks.
> Code Warning
> DB
> org.apache.hadoop.hbase.regionserver.CompactingMemStore.initInmemoryFlushSize(Configuration)
> uses the same code for two branches
> Bug type DB_DUPLICATE_BRANCHES (click for details)
> In class org.apache.hadoop.hbase.regionserver.CompactingMemStore
> In method
> org.apache.hadoop.hbase.regionserver.CompactingMemStore.initInmemoryFlushSize(Configuration)
> At CompactingMemStore.java:[line 140]
> At CompactingMemStore.java:[line 143]
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
