WebSphere Installations
Installations really need to be viewed in two ways:
1. Cells: ./WebSphere/DeploymentManagerN where N is the cell number and 1 may be optional or the undecorated directory may be left for unmanaged cells
2. Nodes: ./WebSPhere/AppServerM where M is the node number on that disk (usually also machine but not necessarily so.
There is no necessary relationship between M and N. A node may stand alone or belong to any cell we wish. Perhaps there needs to be a standard here? No deployment manager on a machine that has a node federated to another cell. It would not allow for the cross coverage of cells on hardware. I’m not sure why I want that,. Here’s a scenario:
Machines A1, A2 and A3 belong to cell C1. Machines B1, B2 and B3 belong to cell C2. The deployment managers for these celss are on nodes A1 and B1. The loss of A1 or B1 denies us management capability of cell C1 or C2.
Assuming web server redundency, I still have 2/3 application coverage for the failed cell and still have failover coverage assuming volume can be accomodated. I can’t add more nodes to the cell without fiddling with the plug-in and other configs manually. If B1, B2 and B3 have secondary nodes federated to A1 and thus members of C1, I have backup capacity to handle load, but can I leave them inactive and start them (from the node agent with wsadmin?) while A1 is unavailable?