We have continually emphasized the need to avoid SPOF in the design of our clusters. We need to carry that philosophy through all aspects of our IT solutions. In particular, I am thinking of:
High available network design for client connections : This can be a difficult task because if users are using PCs to access applications, can the PC operating system support multiple LAN connections? Can that same PC support dynamic routing protocols? And what happens if the PC itself fails? Do we have IT processes in place whereby we can rectify those SPOFs?
Do users log in to application servers ? If so, what solution do we have to allow them to log in to another node in the event of a failure? Have you considered using NIS, NIS+, or LDAP to distribute managing user IDs and passwords?
Are there performance issues when running applications in a cluster ? Have you considered the impact of running multiple applications on a single node. It is good practice to discuss this with user groups and make them aware that during a failure they may experience a drop in overall application performance. Is this acceptable, or will you need to employ more hardware to avoid such performance issues? Have you established Service Level Agreements (SLA) with your user groups relating to performance requirements? Have you considered using PRM and/or WLM to help you achieve those SLAs?
Managing clusters requires this holistic view to high availability. Good luck. See you in the next section.
