Empirical Evaluation of Decentralized Genomic Data Computation Using Bacalhau and IPFS

Large-scale genomic analysis typically relies on centralized infrastructures, creating conflicts between collaborative needs and data sovereignty regulations. This study solves this dilemma by evaluating a decentralized architecture designed to facilitate secure, inter-institutional genomic computation without moving raw data. We integrated Bacalhau for orchestration and IPFS Cluster with CRDT consensus for storage, employing AES-256 encryption. A quantitative evaluation was conducted on AWS using five t3.medium nodes to simulate a resource-constrained hospital network. We tested three scenarios: a centralized baseline (SSH+SCP), an ideal decentralized workflow, and a "chaos" scenario involving active network fault injection. While the centralized baseline was the fastest (Mean=37.69s), the decentralized architecture incurred a manageable ~30% overhead under ideal conditions (Mean=49.22s, SD=1.58s). Critically, under chaos fault injection, although execution time increased to 90.67s (SD=17.84s), the system achieved a superior 100% job completion rate compared to the fragile baseline. This research quantifies the trade-off between execution speed and system resilience in a healthcare context. We demonstrate that this architecture prioritizes data sovereignty and high availability over raw speed, offering a proven model for privacy-critical Decentralized Science (DeSci) collaborations.