Probably the most obvious benefit arising from using multiple writable PostgreSQL nodes is one of reduced write latency. Consider an initial cluster that may resemble this diagram:
Each write to Tokyo or Sydney must first cross thousands of miles before being committed. And due to how replication works, the local replicas in those regions will have to wait for the transaction to be replayed before it will be visible there. These times can be considerable. Consider this table of round-trip-times for network traffic for the locations we've chosen:
|
Dallas |
Sydney |
Tokyo |
|
|
Dallas |
X |
205 ms |
145 ms |
|
Sydney |
200 ... |
