An engineering firm replicates large CAD files daily between two data centers over a constrained MPLS link, sending the same blocks repeatedly with ample memory on the routers. Branch voice and small web flows see no benefit. Which statement correctly describes data redundancy elimination and where it helps most?

Data redundancy elimination builds a symmetric fingerprint dictionary of previously seen byte strings at both WAN endpoints. When the same CAD blocks cross the constrained MPLS link day after day, the routers can send short references instead of the repeated data, reducing WAN use; ample memory supports retaining the history that makes this useful for repetitive bulk transfers. Compressing packets independently can save bandwidth, but it discards history after each packet and cannot recognize matching blocks from earlier transfers. Tuning TCP windows can improve bulk throughput on constrained paths, but it changes delivery behavior without removing duplicate payload bytes. Caching signatures can identify repeats, but small voice payloads and tiny web transactions are less likely to repeat, so they offer few dictionary hits and see little benefit. As an exam caveat, distinguish persistent shared history from one-time compression and TCP window tuning: elimination requires paired devices, sufficient memory, and repeatable plaintext patterns, and does little for encrypted, unique, or interactive traffic. For a practical check, verify that both peers are enabled, monitor cache hit rates and compression ratios during daily replication, and confirm that CAD transfers use less MPLS bandwidth while voice and browsing metrics remain largely unchanged.