Compute and storage decoupled architecture is proven to be feasible in a cloud environment. Almost all cloud vendors develop their own NewSQL base on that architecture. Most of them use InnoDB and write the data to distributed storage. However, the IO system in InnoDB is optimized for local disks. There are many different behaviors between local disk and distributed storage. We need to optimize the IO system to make it run faster.
This presentation will introduce you to the inner workings of the IO module in InnoDB. We will also compare the differences between our own distributed storage system, POLARSTORE, and local disks. Then we will list the optimization points in InnoDB that may influence the performance and show the result after we do the work.
By attending this presentation you will learn how you can optimize your own NewSQL base on "compute and storage decoupled" architecture. You'll also see how to optimize InnoDB on your own distributed storage system since we will list all the IO in InnoDB and tell you the how to attend to these.
Zongzhi Chen is a Staff Engineer in Alibaba Cloud, where he works in the Cloud Database Products Group.
Before joining Alibaba, Zongzhi Chen worked for 7 years in Database and Distributed Storage at Qihoo/Baidu. At Qihoo, he developed a popular large-storage Redis called "pika", which is widely used in Qihoo. Pika is also open source in github and more than 20+ famous and large company in China such as Weibo, Kuaishou use it in their core system. He developed a distributed key-value storage system called zepplin in Qihoo, zepplin is widely used in the Qihoo Search system. He also developed the raft library called "floyd", "floyd" is widely used in many projects in Qihoo.
Zongzhi Chen graduated from Southeast University in China with a bachelor degree and lives in Hangzhou now.