Workshops Archive

FAB-Cache: A Programmable Multi-Site Caching Service for Remote Dataset Retrieval Using FABRIC


Workshop: 13th Annual International Workshop on Innovating the Network for Data Intensive Science — INDIS

Authors: Pinyi Shi, Vaiden Logan, Dakota Morgan, James Griffioen, Zongming Fei, and Kenneth Calvert (University of Kentucky)

Abstract: Scientific datasets, which often range from gigabytes to hundreds of terabytes, are frequently accessed within research institutions and shared among collaborators distributed globally. Thus, efficiently distributing these large-scale datasets to researchers has become a growing challenge. We present FAB-Cache, a programmable, multi-site caching service deployed on NSF FABRIC testbed to accelerate access to remote scientific data. FAB-Cache chains NGINX proxies across geographically distributed FABRIC nodes and caches requested data on demand. It uses perfSONAR throughput measurements to select proxy paths. This design combines distributed storage and network measurements to support a stateful service within the network. We evaluate FAB-Cache using real-world scientific datasets retrieved across multiple FABRIC sites. Cache hits substantially outperform direct downloads over the public Internet, while cache misses achieve comparable or higher throughput in several scenarios. These results demonstrate the potential of FABRIC infrastructure to support flexible data distribution services for scientific workflows.


Back to 13th Annual International Workshop on Innovating the Network for Data Intensive Science — INDIS Archive Listing

Back to Full Workshop Archive Listing