Move Big Data solutions based on Percona XTRADB cluster

move Big Data solutions based on Percona XTRADB cluster      a mobile trip to the IoE     

Recently, because of the "prism door" incident exposure, caused the national attention to information security issues, the major industries also launched to "IOE" action. For mobile, on the one hand is the concern about information security, on the other hand, to reduce cost considerations, the introduction of open source architecture has become a realistic solution.

In the Internet industry, the use of MySQL has become mainstream, but as Oracle's takeover of Sun, MySQL's control falls into the hands of Oracle, the risk of a possible closed source of MySQL has become an industry consensus. Thus, a variety of MySQL branches are produced. This paper mainly analyzes the PERCONAXTRADB as a database scheme.

Second,Percona stradb cluster analysis
Website:http://www.percona.com/

Percona Xtradbcluster is a high availability and scalable solution for MySQL.

Percona Xtradbcluster provides features such as:

1. Synchronous replication, transactions are either committed or not committed at all nodes.

2. Multi-master replication, you can write on any node.

3. In parallel application of events from the server, the true meaning of parallel replication.

4. Automatic node configuration.

5. Data consistency, which is no longer asynchronous replication.

Percona Xtradbcluster is fully compatible with MySQL and Percona Server and is represented in:

1. Compatibility of data

2. Application compatibility: No need to change the application

Cluster Features:

? A cluster is made up of nodes, and it is recommended to configure at least 3 nodes, but it can also run on 2 nodes.

? Each node is a normal Mysql/percona server, you can make the existing database server into a cluster, conversely, you can also split the cluster into separate servers.

? Each node contains a complete copy of the data.

The advantages are as follows:

1. Executes on the local node when a query is executed. Because all data is local, no remote access is required.

2. No need for centralized management. Any node can be lost at any point in time, but the cluster will work as usual.

3. Good read load extension, any node can be queried.

Disadvantages are as follows:

1. Adding a new node is a big expense. You need to copy the full data.

2. Unable to effectively resolve the write scaling problem, all write operations will occur on all nodes.

3. The number of nodes there is the number of duplicate data.

The architecture diagram is as follows:

The difference between Percona XtraDB cluster and MySQL replication is:

Cap theory for Distributed systems:

C-Consistency, the data of all nodes is consistent;

A-availability, one or more nodes fail, does not affect the service request;

P-Partition tolerance, the connection between nodes is invalid, can still process the request;

Any distributed system needs to meet the three of the two.

Mysqlreplication: Availability and zoning tolerance;

Percona Xtradbcluster: Consistency and availability.

Therefore, MySQL replication does not guarantee data consistency, while Percona XtraDB cluster provides data consistency.

Percona Xtradbcluster Components:

Percona XtraDB Cluster XtraDB-based perconaserver and contains write-replica set patches. Use Galera 2.xlibrary, a common multi-master synchronous replication plug-in for transactional applications.

The new features of Galera 2.x are:

1.IST (incrementalstate Transfer) incremental state transfer. is particularly useful for WANs.

2.RSU (rollingschema Update) rotation update schema. Operations on the table are not prevented.

third, the system architecture based on Percona stradb cluster       Extensibility Architecture   

Vertical shards:

Multiple Percona XTRADB cluster;->spring Configure multiple data sources;

Horizontal shards:

Multiple perconaxtradb cluster;->guzz for data horizontal slicing and ORM;
lvs+keepalived Extensions: the equivalent routing technology with three layer switch can configure up to 8 LVS servers, each of which is master,lvs configured as Dr Mode, which eliminates the LVS bottleneck.
v. Summarywe are prepared to use the above scheme in the mobile Internet project, which is calculated in 50 million PV, 10000 concurrent cases, need to 1,828,571tpmc server one. Press 8 LVS Server, each forwarding 100 Percona xtradb cluster, each cluster 8 server compute (the cluster each node data same), equivalent to 8x100=800 server. Each server tpmc by 1,807,347, support 50 million PV calculation, that is 800x5000 = 40 billion PV. Sufficient to meet the needs of the project.      The above scenarios also need to be validated in subsequent ongoing operations. Please criticize treatise!