TWJUG 201909 聚會

FB社群: https://www.facebook.com/groups/185338705012/

TWJUG Slack: https://bit.ly/2DY6Lhj

贊助單位:

  • 感謝 LINE 贊助本次活動的場地和食物!
  • 感謝 JetBrains 贊助個人版授權一年份 x 1!

活動內容:

一、

主題: ZGC for Future LINE HBase    

講者: Shinya Yoshida

Shinya Yoshida is a Serverside Engineer for LINE. He is working for Dev/Ops of HBase clusters for LINE Messaging Service. He was contributed to OpenJDK Projects like Lambda and Kulla, and is now a committer of OpenJDK and a board member of JJUG(Japan JUG).

簡介:

LINE Messaging service uses HBase as a persistent store, which is running on JVM. GC pause affects HBase response time and the service finally. We're using G1GC for our HBase, but we still have many 100+ ms pauses. And GC pause time will be a bottleneck of increasing JVM heap size to get the full-power of HBase because if we increase the heap size, the pause time is getting more longer.

At Java 11, they introduced extremely low pause time GC as an experimental feature, ZGC. We considered ZGC will free us from HBase GC pause issues and evaluated it with HBase in our test cluster.

I show the basis of GC and which part is different from other GCs and ZGC. I also share the points to choose GC and why we considered ZGC fits our HBase. And I show the evaluation results of ZGC with HBase. I also talk future plan to apply ZGC to our production environment.

二、

主題: Kafka Broker performance degradation by mysterious JVM pause    

講者: Yuto Kawamura

Yuto Kawamura is a Senior Software Engineer at LINE.

He's leading a team providing company-wide Kafka platform and working for a lot of reliability/performance engineerings.
He's an Apache Kafka contributor, also he gave a lot of presentations in large conferences such as Kafka Summit 2017/2018, LINE DEVELOPER DAY 2018 and he loves visiting 九份.

簡介:

    At LINE we are operating one of the largest scale Kafka cluster operated as a middleware infra for many services. Since it receives large scale traffic such that over 260 billion messages / day, we often meet a performance issue and try to deal with it. Recently we had an interesting case which makes Kafka broker to lowly perform due to insanely long JVM pause which wasn't caused by GC nor other typical causes. In this talk I'm gonna talk how did we investigated this issue diving beyond  into the Linux kernel and finally solved it.

注意事項: 

• 活動只需要在這邊或是 meetup.com 其中一邊報名就好,請勿兩邊報名喔 :)
 

 
LINE 台灣辦公室 / 114台北市內湖區瑞光路610號 10樓

Event Tickets

Ticket Type Sale Period Price
一般票

2019/08/19 12:00(+0800) ~ 2019/09/05 18:00(+0800) End of Sale
  • Free
Next Step