| Multiple BYNETs |
Multinode database servers are equipped with at least two BYNETs. Interprocessor traffic is never stopped unless all BYNETs fail. Within a BYNET, traffic can often be rerouted around failed components. |
| RAID disk units |
- Nonarray storage cannot use RAID technology.
- Database servers use Redundant Arrays of Independent Disks (RAIDs) configured for use as RAID1, RAID5, or RAIDS.
- RAID1 arrays offer mirroring, which maintains identical copies of data.
- RAID5 or RAIDS protects data from single-disk failures with a 25% increase in disk storage to provide parity.
- RAID1 provides better performance and data protection than RAID5 or RAIDS, but is more expensive.
|
| Multiple client-server connections |
In a client-server environment, multiple connections between mainframe and workstation-based clients make sure that most processing continues even if one or several connections between the clients and server are not working. Vproc migration is a software feature supporting this hardware issue.
|
| Isolation from client hardware defects |
In a client-server environment, a server is isolated from many client hardware defects and can continue processing despite such defects. |
| Power supplies and fans |
Each cabinet in a configuration has redundant power supplies and fans to make sure of fail-safe operation. |
| Hot swap capability for node components |
Allows some components to be removed and replaced while the system is running. Vantage offers hot swap capability for the following: - Disks within RAID arrays
- Fans
- Power supplies
|
| Cliques |
- A clique is a group of nodes sharing access to the same disk arrays. The nodes and disks are interconnected through FC buses and each node can communicate directly to all disks. This architecture provides and balances data availability in case of a node failure.
- A clique supports the migration of vprocs following a node failure. If a node in a clique fails, its vprocs migrate to another node in the clique and continue to operate while recovery occurs on their home node. Migration minimizes the performance impact on the system.
- PEs that manage TPA-hosted physical channel connections cannot migrate, because they depend on the hardware that is physically attached to the assigned node.
- PEs for workstation-attached connections do migrate when a node failure occurs, as do all AMP vprocs.
- For maximum fault tolerance, put no more than one node in a clique in the same cabinet. Usually the battery backup feature makes this precaution unnecessary.
|