Xinnor and the University of Utah’s SCI Institute Replace Aging PostgreSQL Backend with xiRAID

Key Takeaways

• xiRAID delivers 39× higher read IOPS than legacy single‑drive setup
• RAID10 on ten NVMe drives reaches 8.24 million read IOPS
• PostgreSQL 17 on xiRAID achieves 2.65× throughput boost
• Operational setup completed in minutes, far faster than mdadm

Pulse Analysis

Research institutions that manage billions of files across petabytes of storage face a unique challenge: metadata databases must keep pace with ever‑growing catalogs. The SCI Institute’s previous PostgreSQL deployment ran on a single NVMe drive, causing query latency to spike to 24 ms and throttling data‑intensive workflows. By moving to a purpose‑built RAID solution, the institute not only eliminated the storage bottleneck but also future‑proofed its infrastructure for continued data growth, a concern shared by many university labs and national labs alike.

xiRAID’s architecture leverages a lock‑less read path and a RAID10 layout across ten 3.84 TB enterprise NVMe drives. Benchmarks show the raw hardware ceiling at 8.03 million read IOPS, while xiRAID pushes that to 8.24 million IOPS with 129 GB/s throughput—only a 6 % overhead when layered with ext4. In production, PostgreSQL 17 on xiRAID processes 15,069 tps at 0.265 ms latency, a 2.65× improvement over the legacy system. These figures illustrate that high‑performance RAID can match or exceed raw device limits while maintaining data integrity and reliability.

Beyond raw numbers, the deployment underscores the business value of operational simplicity. xiRAID’s initialization completed in minutes, a stark contrast to the days‑long setup typical of mdadm or ZFS. The single‑command lifecycle management and responsive vendor support reduce administrative overhead, allowing IT teams to focus on research outcomes rather than storage mechanics. As data‑intensive fields such as genomics, climate modeling, and AI continue to expand, solutions that combine enterprise‑grade performance with ease of use are poised to become the new standard in scientific computing environments.