Questions about ES Performance optimization, can anyone help me?

[xiaodong.hu (hu xiaodong)]

Hello, I'm testing indexing performance about es.

under these conditions :

1. ES 5.4.0 , a single node cluster.

2. -Xmx17179869184 -XX:MaxDirectMemorySize=17179869184

3. use _bulk api, 500docs per request.

4. 20 clients of jmeter.

5. path.data:
   /data2/zdh/elasticsearch/data,/data3/zdh/elasticsearch/data,/data4/zdh/elasticsearch/data,/data5/zdh/elasticsearch/data,/data6/zdh/elasticsearch/data,/data7/zdh/elasticsearch/data,/data8/zdh/elasticsearch/data,/data9/zdh/elasticsearch/data,/data10/zdh/elasticsearch/data,/data11/zdh/elasticsearch/data

6. index named test with 12 shards.

7. disk infos.
   

8. memory info
   

9. cpu about 48 processors, top info:
   

results:
I use Jconsole to watch es, and I found lots of threads blocked at TranslogWriter.add()

and I found the memory used by off-heap is only 1G:

Is that OK?

and the IO:

at one moment , just one or two hard disk are busy, the others are idle. I do't know why.

thread pool about bulk:

"thread_pool" : {
"bulk" : {
"threads" : 32,
"queue" : 88,
"active" : 32,
"rejected" : 325,
"largest" : 32,
"completed" : 165387
}

memery usage of jvm: 37%

I think the disk is not been fully used , Is there anything I can do to optimize the performance? thank you.