MySQLrockdb Performance_Using TerarkDB to Improve MySQL Performance.pdf

Using TerarkDB to improve the performance of MyRocks

——MySQL performance optimization in random read scenarios

Lei Peng@

CTO

Outline

l Introduction to MyRocks

l Why MyRocks

l Block Compression for Traditional Databases

l Hardware Development Trend

l The relationship between TerarkDB and MyRocks

l The principle, advantages and disadvantages of TerarkDB

l New problems， and optimization suggestions

Introduction to MyRocks

l MyRocks=MySQL+RocksDB (storage engine)

l Based on MySQL-5.6 ： code (MySQL onTerarkDB)

Ø Added some MySQL-5.7 Features

Ø Added some syntax and functions specific to RocksDB

Ø Some functions do not support，details See ：MyRocks-limitations

l Learn more about MyRocks ：MyRocks DeepDive

MyRocks main functions

l InnoDB-like functions

l Transaction

Ø Based on MVCC

Ø ReadCommitted, RepeatableRead(PostgreSQL-style)

l Online Backup

Ø Logical backup (using mysqldump)

Ø Binary backup (using myrocks_hotbackup)

Why choose MyRocks

l mature and stable，meets most functional requirements

l Use a single DB， suitable for cloud DB multi-tenant scenarios

l Based on RocksDB， directly compatible with TerarkDB

l Based on LSM，Yes SSD friendly

Traditional databases all use block compression——

l InnoDB

l LevelDB/RocksDB…

l TukoDB

Block compression has an insurmountable ，

l WiredTiger inherent flaw ，

l Cassandra/ScyllaDB let’s see how it works& #xff0c;

l CouchDB analysis –

l HBase

l …

Principle of block compression

l Multiple pieces of data are packed into one block ， the index locates the block

Ø Increase the size of the block ： the index is smaller ， the number of data pieces in a single block is more ， invalid decompression more

Ø reduced block ： index larger ， compression algorithm not working efficiently ， less compression

l Block cache DirectIO leads to only

problem 1

decompressed data ， a certain process

too slow——

Ø Decompress a block for each access ，

Reduced memory utilization to a certain extent

Ø Put the decompressed block into the cache

p>Problem 2 DirectIO requires aligned access ，

Ø The operating system also has a file cache double-buffer problem！

Using DirectIO requires cross-boundary in the worst case

(compressed data is stored in the file) Read two more blocks

The development trend of hardware

l The capacity/price of the mechanical hard disk still has a huge advantage

Ø More resistant to writing，safer；helium hard drive，HAMR technology…

l SSD is getting cheaper，NVMe/ PCIe has been widely used

Ø It is not resistant to writing ， If the data is damaged, it will be permanently lost

Ø SSD with NVMe/PCIe interface is faster，lower latency

l 3DXPoint ， is faster than SSD， more resistant to writing

Ø The first generation product Optane (Flash Teng) is already on the market！

Ø Using DIMM interface ， can be used as memory…

Development trend of hardware

Bandwidth (MB/s) IOPS Latency (us) Access granularity