We have a client that's after high availability infrastructure. They run business critical applications on their app server (ACE currently, which we intend to migrate to AF analytics calculations). This is straightforward enough for PI data archive with a collective. However for AF, my understanding is that OSIsoft recommend having two duplicate AF servers and a load balancer appliance in front of them. The AF servers should point to a SQL cluster.
To provide full HA though, the load balancing would need to be done by two appliances as well, correct? These (virtual) servers will be split across two data centres in different locations about 100km apart, connected with a 100 Mbps WAN link I believe.
Also, if we're using load balancers do we need to configure the ACE schedulers for HA? Or would the load balancing be done at the server level, so that would be taken care of?
Since we're protecting against hardware failure, the load balancers would need to be on different hosts. We'd also need to ensure the SQL cluster is across multiple hosts/nodes.
Is this setup in line with OSIsoft recommendations? Does anyone else have experience with this kind of HA design? Any suggestions on how to size load balancers in terms of number of AF elements/calculations or some other metric?