Views: 1913
Last Modified: 15.09.2022

Select a suitable variant

Before introducing updates you need to patiently analyse (predict) the type of load on the web application. It's recommended to proactively use the known monitoring tools, such as munin, zabbix, apache server-status, etc.

Below are web cluster configuration variants for various types of load.

High load to processors, low/medium load to DBMS, rare content updates

You need to reduce the load on processors - connect the node(-s) to web cluster. Include nodes into load balancer. Result: processor load is parallelized between nodes.

It's also recommended to start a memcached server at each node - this will allow to reduce the load on processors even more due to cache being created once at a single web cluster node, but employed by the rest of the nodes.

For content synchronization use csync2, nfs/cifs-server at one of web cluster nodes.

Increasing high load to DBMS

To remove excessive DBMS load you need to organize a master-slave replication. The more slave nodes are added, the the lower the load on the main master database. If slave nodes have various power, it's recommended to adequately distribute the load between them using the parameter Load, % (0..100).

As the result the load on DBMS will be distributed between master and slave web cluster nodes. WIth further increase of load on the DBMS, it's recommended:
  1. Optimize master and slave nodes. Because the master server will be used predominantly for writing the data, you can configure the parameters appropriately and optimize the server for writing, and optimize slave nodes for data reading.
  2. Add slave nodes.

Large cache volume, frequent cache restructuring, content updates, with significant changes

Your objective is to effectively configure the cache handling. For this purpose, launch a memcached server on each node and allocate several GBs of memory for each (depending on the application requirements).

For content synchronization, it's recommended to use a dedicated nfs/cifs-server, ocfs/gfs or equivalent.


0


Courses developed by Bitrix24