[
https://issues.apache.org/jira/browse/HDFS-8134?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yingqi Lu updated HDFS-8134:
----------------------------
Attachment: pic2.png
pic1.png
Hi All,
Here are the most recent data on OpenJDK with HDFS.
Purpose of the study:
The goal is to show OpenJDK is enterprise ready from the performance point of
view.
Configuration:
• Intel Xeon® E5-2699 V3 (2 X 18Core 2.3Ghz) CPUs
• BIOS Version: SE5C610.86B.01.01.0008.021120151325 (release date
02/11/2015)
• All BIOS settings are kept default (HT enabled, Turbo enabled, Power
features enabled)
• Memory: 16 X 16GB DDR4 2133MHz, 2 Dimms per channel
• Storage: OS is installed on a 120GB SSD. HDFS and tmp directory is
located on 1 PCIeSSD drive (1 X Intel® SSD DC P3700 Series, 1/2 Height PCIe
3.0, 20nm, MLC)
• OS: CentOS 7 kernel version 3.10.0-123.el7.x86_64
• Hadoop: 3.0.0-SNAPSHOT (commit
867d5d2675b8fb73c40fac1e581b02b005459d95, dated 04/02/2015), single node cluster
• Java: Java HotSpot(TM) 64-Bit Server VM (build 25.40-b25, mixed mode)
vs. OpenJDK 64-Bit Server VM (build 25.40-b25, mixed mode). Two GC methods are
checked – G1GC and ParallelOldGC.
• Workload: dfsioe inside HiBench suite. Data size is 128M. Read and
write operations are included in the performance data study.
Data summary:
1. With 128M file size and 1000 files, both read and write operations from
dfsioe show similar performance (throughput) between Hotspot Java and OpenJDK
(5% performance difference is within workload run-run variance range).
2. We also tested 2 GC methods – G1Gc and ParallelOldGC. They both show
similar performance as well for this specific workload.
3. The single system cluster runs 95%+ CPU utilization for both read and
write operations.
Performance charts are attached here. Please let me know if you have any
questions and comments.
Thanks,
Yingqi Lu
> Using OpenJDK on HDFS
> ---------------------
>
> Key: HDFS-8134
> URL: https://issues.apache.org/jira/browse/HDFS-8134
> Project: Hadoop HDFS
> Issue Type: Task
> Components: benchmarks, performance
> Environment: CentOS7, OpenJDK8 update 40, Oracle JDK8 update 40
> Reporter: Yingqi Lu
> Assignee: Yingqi Lu
> Attachments: pic1.png, pic2.png
>
>
> Dear All,
> We would like to start the effort of certifying OpenJDK with HDFS. The effort
> includes compiling HDFS source code with OpenJDK and reporting issues if
> there is any, and completing performance study and comparing all the results
> with Oracle JDK. The workload we will start with is DFSIOe which is part of
> the HiBench suite. We can surely add more workloads such as Teragen and etc.
> into our testing environment if there is any interest from this community.
> This is our first time to work on this community. Please do let us know your
> feedback and comments. If you all like the idea and this is the right place
> to start the effort, we will be sending out the data soon!
> Thanks,
> Yingqi
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
