Mailing-List: contact commits-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@cassandra.apache.org
Date: Tue, 7 Jul 2015 12:48:04 +0000 (UTC)
From: "Roman Bielik (JIRA)" <jira@apache.org>
To: commits@cassandra.apache.org
Message-ID: <JIRA.12843175.1436273238000.116904.1436273284563@Atlassian.JIRA>
In-Reply-To: <JIRA.12843175.1436273238000@Atlassian.JIRA>
References: <JIRA.12843175.1436273238000@Atlassian.JIRA>
 <JIRA.12843175.1436273238063@arcas>
Subject: [jira] [Created] (CASSANDRA-9747) JNA makes Cassandra slower
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit

Roman Bielik created CASSANDRA-9747:
---------------------------------------

             Summary: JNA makes Cassandra slower
                 Key: CASSANDRA-9747
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-9747
             Project: Cassandra
          Issue Type: Bug
         Environment: Cassandra 2.0.14; JNA 4.x; 32-core CPU; 64GB RAM (8GB heap); 2 physical HDDs used as commit and data; Java 1.7
            Reporter: Roman Bielik
            Priority: Minor


When JNA is loaded (as recommended in the production settings), it has negative effect on performance, especially when used with HDD => it significantly increases the disk IO wait. 
It looks like JNA works fine only when used in combination with SSD disk. Otherwise it is better not to use JNA at all. This is very unexpected behavior, can you please analyse?

Test description:
Roughly 50% reads and 50% writes of about 1.5kB records using Thrift interface.

Test results (sorted by performance):

*HDD + JNA*
Worst performance. High IO wait.

*HDD, no JNA*
Good performance. Low IO wait, but the Cassandra CPU usage is quite high.

*SSD + JNA*
Seems to provide the best performance. Low IO wait. Low Cassandra CPU usage.


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)