Below is a list of basic recommendations that can be used on creating a BI servers cluster:
Any available balancer can be used if it meets the following requirement: the balancer must provide linking queries of certain client to the certain server.
As a state server, Foresight Analytics Platform supports work with Redis and Memcached.
Number of BI Servers. The number of BI servers started within one OS is determined by the formula: Number of Users/300, and the result is rounded up. For example, if 700 users are expected to work, it is recommended to start three BI servers. Generally, it is recommended to use at least two BI servers to be able to redirect queries to another server in case of the first BI server breakdown.
Number of Operational Systems that Run BI Servers. The number of operating systems is determined from 500 users to one OS. For example, if 1000 users are expected to work, it is recommended to start two virtual machines, each of them contains two BI servers.
Recommendations for the number of BI servers and operating systems can be applied if the response time is short (1-4 seconds). If the developed web application shows long response time and suffers from the lack of processor time, the number of working users will not have much influence and can be increased.
The specified recommendations are of advisory nature; anyway, the final version of the cluster can be accepted after personal load test for the specific developed web application. Resulting characteristics of the cluster are also influenced by such factors, as intended load, intended peak load or increasing fault tolerance requirements. The number of started BI servers or OS should not be unreasonably increased because it may cause loss of performance due to excessive resource usage to deploy cache, memory used by processes related to OS, and so on.
See also: