[
https://issues.apache.org/jira/browse/ZOOKEEPER-2835?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16113781#comment-16113781
]
ASF GitHub Bot commented on ZOOKEEPER-2835:
-------------------------------------------
Github user asdf2014 commented on the issue:
https://github.com/apache/zookeeper/pull/301
Hi @afine . Thanks for your comments. Your consider is entirely correct.
But I found the [PR#16937](https://github.com/elastic/elasticsearch/pull/16937)
from `elasticsearch`. And every node in our `elasticsearch` cluster always be
set as huge
[`ES_HEAP_SIZE`](https://www.elastic.co/guide/en/elasticsearch/guide/current/heap-sizing.html)
value. Then it just spend a few second when the `elasticsearch` instance start
up. However, the server node in `zookeeper` cluster usually we don't assign too
large memory. So, the `zookeeper` as daemon instance will get more benefit from
it. (PS, same tuning in
[Cassandra](https://tobert.github.io/tldr/cassandra-java-huge-pages.html),
[Jenkins](https://jenkins.io/blog/2016/11/21/gc-tuning/) etc.)
> Run server with `-XX:+AlwaysPreTouch` jvm flag
> ----------------------------------------------
>
> Key: ZOOKEEPER-2835
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-2835
> Project: ZooKeeper
> Issue Type: Improvement
> Components: server
> Affects Versions: 3.5.3
> Reporter: Benedict Jin
> Assignee: Benedict Jin
> Original Estimate: 72h
> Remaining Estimate: 72h
>
> Add `-XX:+AlwaysPreTouch` jvm flag for server, let jvm through demand-zeroed
> way to allocate memory once in place when the process started
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
