A major financial firm operates the largest message switch in the U.S., with hundreds of thousands of attached devices. It needed to coordinate multiple HPE NonStop nodes in an active/active configuration. These nodes were distributed around the U.S. for disaster tolerance.
HPE Shadowbase Solution: Homogeneous Active/Active Systems
Collision identification and resolution via data content are illustrated in Figure 1. Both ATM and POS systems’ transactions may be routed to either Node A or Node B in an active/active configuration. As a request is received, it is applied and the application updates the record’s “update-timestamp” field, and then the change is asynchronously replicated to the other node so that all copies of the database are kept up-to-date. When data collisions occur (due to the nature of asynchronous replication), they are resolved by a predetermined algorithm, which looks at a request’s record contents and applies the request with the most recent update timestamp (the other change loses and is logged for further review).
- The system supports true load-balancing by assigning each new request to the least heavily loaded node.
- Since the databases are kept synchronized via asynchronous replication, data collisions can occur. However, when the replication engine detects a collision, the engine resolves it automatically by using the data content of the record updates to determine which update to accept and which to reject.
- For example, if the records have a time-stamp field indicating the time of the update, the replication engine can be directed to keep the most recent update and log the rejected update for later analysis, if desired. Otherwise, the rejected update can simply be discarded. Other algorithms may also be used, such as choosing the highest or lowest value of a specified field, or choosing the earlier update, depending on the underlying application’s need.