Geo-Replicated File Systems at Near-Local Speeds: Evaluating DAOS on the All-Photonic Network

Geo-replicated distributed file systems offer a convenient solution to avoid data loss caused by disasters or system failures. However, ensuring reliable data preservation at high I/O performance is challenging because it requires synchronous replication, which acknowledges write completion only after the replication is completed. This process makes its performance highly susceptible to network latency. To mitigate this issue, we leverage the All-Photonic Network (APN), which enables low-latency wide-area communication. In this paper, we evaluate the performance of the Distributed Asynchronous Object Storage (DAOS) file system in an APN-based experimental environment. Even with synchronous replication to servers 50 km away, experimental results demonstrate high I/O performance better than or comparable to local SATA SSDs in realistic workloads.