why i can't configure "-XX:+UseConcMarkSweepGC" in hbase-0.94.6-cdh4.4.0

[#1]

Hi all,

I read the book "hbase in action" and check the configuration for hbase gc. But when i use the same configuration in hbase-0.94.6-cdh4.4.0, and find that :"

Unrecognized VM option 'UseConcMarkSweepGC'
Could not create the Java virtual machine.

"

why these messages happen?

Another questions : 

why some scripts are disappeared in cdh4 hbase like "add_table.rb" and "graceful_stop.sh"?
did Cloudera aggregate or deprecate them?

And some one advice me that "config 'hbase.regionserver.hlog.blocksize * hbase.regionserver.maxlogs' larger than 'hbase.regionserver.global.memstore.lowerLimit * HBASE_HEAPSIZE'."  Is it correct for hbase cdh4.4.0?

I find that 'hbase.regionserver.hlog.blocksize' and 'hbase.regionserver.maxlogs'  are not the default properties in hbase-default.xml. So can you tell me the default values of these properties. 

Thanks for your reply！

yyx,

[#2 · Accepted solution]

@yyx for your GC settings question: this is not a factor of CDH at all.  This is something that your JVM controls.  What JDK are you using?

2nd Question: I believe this thread on our mailing list solves it.

Your final questions about the property settings are a bit confusing to me.  I'm not seeing how the Hlog settings relate to lowerLimit of memstore, but I can tell you what they default to in a Cloudera Manager cluster:

hlog.blocksize always defaults to your HDFS block size, unless you override it.

maxlogs defaults to 32

lowerLimit defaults to .35

upperLimit defaults to 0.4

HTH

[#3]

Hi clint,

Thank you for your help.

I use JDK6u45. And now i knew that cloudera advice me another script to replcae "graceful_stop.sh".

About the last question, i just found a lot of blogs about "how to improve hbase performance".

There is an advice in these articles : "hlog.blocksize*maxlogs  must be  larger than HBASE_HEAPSIZE*lowerLimit ". So i ask you to varify these advice.

yyx