Figure 4.10 Storage Replica placement in the filesystem stack
The way that Storage Replica works is through log files, and not directly to the final
storage, as shown in Figure 4.11. With Storage Replica, both the source and
destination have a log file that is the initial target for all writes. With synchronous
replication, it works as follows:
1 . The write originates from an application.
2 . The write is hardened to the log file on the source and sent to the replica server.
3 . The write is hardened to the log file on the destination (replica) server.
4 . The write is acknowledged to the source server.
5 . The write is acknowledged to the originating application now that it has been
persisted to both servers.
6 . At some point in the future, the write is lazily written from the log to the data disk
(although this is often within milliseconds).