Oracle's high availability and load balancing

Browse through Oracle's official web pages and unofficial PPT to get a quick overview of the high-availability scenarios Oracle offers.
1. RAC
RAC, Real application Clusters
Multiple Oracle servers form a shared cache, and these Oracle servers share a network-based storage. This system can tolerate single-or multiple-machine failures.
However, many nodes inside the system need high-speed network interconnection, basically is to put all things in a room, or a data center. If the engine room fails, such as the network is not working, then it is bad. So the only use of RAC or can not meet the general Internet company's important business needs, important business needs a multi-engine room to tolerate the accident of a single room.

2. Data Guard.
The Data Guard scheme is suitable for multi-engine rooms. A computer room a production database, and other computer room to deploy standby database. The standby database is divided into physical and logical. The physical standby database is mainly used for switching after production failure. The logical standby database can share the read load of the production database in peacetime.

3. MAA
MAA (Maximum availability Architecture) is not an independent third, but a combination of the previous two, to provide the highest availability. RAC clusters are deployed in each room, and data guard is used to synchronize between multiple rooms.

High-availability solutions for Oracle+fusionio+dataguard
Traditional Oracle's high-availability scenarios must be based on shared storage devices, whether it is a dual-machine master-Standby mode or an Oracle RAC, and the database must be placed on a shared SAN storage with HA or cluster software for high availability. Oracle Dataguard is a good disaster recovery software, but as an HA solution, features have many limitations, such as data loss, application transparent switching, read-only unreadable (11g) and so on, there is no very good solution at present.
Since the advent of solid state storage technology, the single-machine IO capability has greatly improved, such as the Fusionio card with PCIe interface, the single-block card can provide tens of thousands of IOPS capacity, the single-machine IO capability has exceeded the traditional disk storage. However, what has been bothering us is how to solve the high availability of Oracle databases in a shared-free storage environment? Our team has designed a framework that provides simple, reliable, and highly available functionality.

While in Oracle's position it is always advisable for customers to be better able to plan their own applications, when there are other load balancing methods, try not to rely on the Oracle load balance method, but often after configuring the Oracle RAC database to the customer, Customers will be asked to test load balancing (Balance) and TAF (Transparent application Failover), and use both tests as a standard for the success of the RAC installation.
This is a very helpless thing, like the side of the stub as the main function, and even some maiduhuanzhu feel, but after all, this is the customer, more understanding of Oracle Load Balance (after the use of lb), to better meet customer needs.

The load balancing of RAC is primarily when a new session is connected to a RAC database, and how to determine which node the connection is going to work on. In RAC, load balancing is divided into two types, one based on the client connection and the other on the server side. The load balancer configuration of the client is relatively simple, just add load_balance=on as an option in Tnsnames.ora.

Load Balancing (Balance) is one of the most important features of Oracle RAC, which distributes the load evenly across nodes in the cluster to improve the overall throughput capacity of the system.
Typically there are two ways to achieve load balancing
One is load balancing based on client connection
Client-side load balancing is primarily achieved by adding Load_balance=yes entries to the Tnsnames.ora
If Load_balance=yes is not turned on, Oracle Net chooses one to connect based on the address list until the connection is successful. If the first host fails to connect, in the case of multiple addresses, the second address connection is then selected, and so on until the connection succeeds. When Load_balance=yes is turned on, Oracle Net chooses an address randomly from multiple addresses to connect until the connection is successful. Note that this connection is only randomly selected based on the address list, regardless of the number of actual connections currently on each instance, or regardless of the actual connection load for each node

The second is based on the information gathered by the server-side listener (Listener) to allocate the new connection request to a less-connected instance of the implementation.
The Oracle RAC server-side load balancer allocates new connection requests to the least-loaded nodes based on the number of connection loads for each node in the RAC. When the database is running, the Pmon process for each node in the RAC updates the connection load of the respective node to service_register every 3 seconds. In the case of any listener failure in the node or unexpected failure of the listener, the Pmon process checks the current node for a reboot every 1 seconds to obtain the latest load information to adjust load balancing in a timely manner.