The size and number of nodes is the determinant of your cluster design. Based on your expected load, you need to specify enough processing power to meet your performance and availability requirements. This includes hardware specifications such as processors and memory, or their equivalent for virtual servers.
One of the design decisions you must make for hardware is whether to have few instances on powerful hardware or many instances on cheaper hardware. If high availability is a key requirement, having more instances decreases the risk and impact of any one failure. Other issues such as maintenance and cost must also be factored into this decision.
Hardware availability can be another issue. Does your budget include the new servers, either real or virtual, to handle the loads you expect? Or can you reuse existing hardware. Remember that having mismatched hardware is possible in the cluster, but it complicates the server configuration and may lead to sub-optimal load-balancing.
Because the node architecture is the key to cluster, running tests with various options can help you choose the right hardware. By performing load tests, you can determine how many users can run on a single node and scale the number of nodes accordingly. You can also run tests with various numbers of processors and memory to determine what configuration is optimal for your expected user load. If you plan on using virtual servers, performing load tests can help uncover any issues with connectivity and stability.
Recommended Comments
There are no comments to display.