Why Riak clusters should be at least five nodes or more

This article link: http://blog.csdn.net/ddjohn/article/details/43523983

First make sure you install Riak correctly, when you use Riak fast track to quickly create 3 nodes in a development or test environment, we recommend that you deploy at least 5 nodes in all production environments to ensure that you benefit from the Riak high availability principle.

Deploying five or more nodes will provide the best performance and increase as the cluster expands. Because the Riak is scaled horizontally by adding nodes, you will find that as the cluster grows larger, performance, reliability, and throughput will become better. Smaller deployments can compromise system fault tolerance and availability: "Robust" Replication requirements (we default to three), in smaller clusters, a single point of failure means that the demand for replication is likely to be unmet. This results in performance degradation and increased risk of data loss. In addition, a cluster of less than 5 nodes will cause 75% to 100% nodes to respond to each request, which will result in an unnecessary load that degrades performance.

Let's take a closer look at the situation of three nodes and four nodes.

Fault tolerance and performance under 3-node clusters:

To ensure that the cluster always responds to read-write requests, Riak recommends a rational replication strategy: Keep a three-point copy on three different nodes. The default configuration for Riak is to require at least four nodes to ensure that there is not a single point to save multiple copies of any particular piece of data. However, this is entirely possible by changing the settings to ensure that three replicas exist in a separate node in a three-node cluster. In addition, you may encounter a single point of failure or a network failure caused by a copy placement problem.

When a three-node cluster encounters a node effect or a network partition, the default replication requirement is still at least three nodes to complete but at this point only two or one node is working. This results in performance degradation and a risk of data loss.

Fault tolerance and performance under 4-node clusters:

Because replication requires 3 nodes, in a cluster of 4 nodes, each request for a specific data needs to be responded to from 75% to 100% nodes, which may degrade the performance of the cluster. If you encounter a single point of failure or a network partition, you will encounter the problem we mentioned above.

What if I change the default replication configuration?

If you want to use the correct number of replicas in your project, just make sure that you use N+2 nodes in your cluster (n is the number of replicas for the above reasons).

If there are 5 nodes:

When your Riak cluster node is increased to 5, the percentage of each node in the cluster that responds to requests drops. Riak doing scale-out is based on this. When encountering a single point of failure, the number of other nodes is large enough to ensure that your data is not lost.

So when you use a cluster in a test or development environment, you can use a smaller number of nodes, but when in a production environment, the number of nodes must be greater than or equal to 5.

English Link: http://basho.com/why-your-riak-cluster-should-have-at-least-five-nodes/

Why Riak clusters should be at least five nodes or more