AI Research
Apr 7, 2026
MIT researchers develop system to enhance data center storage performance
Apr 7, 2026
AI Summary
Researchers at MIT have created a software system called Sandook that improves the performance of storage devices in data centers by addressing three sources of performance variability. This system can nearly double the efficiency of existing storage devices without requiring specialized hardware, thus maximizing resource longevity and reducing waste.
- The efficiency of data centers often suffers from underutilized storage capacity due to performance variability among devices.
- MIT researchers developed Sandook, a software-based system that simultaneously addresses three major sources of performance variability in storage devices.
- Sandook employs a two-tier architecture with a central controller for overall task distribution and local controllers for real-time data rerouting.
- The system adapts to changing workloads and has been tested on tasks like AI model training and image compression, achieving performance improvements of up to 94% compared to traditional methods.
- Sandook profiles the performance of each storage device and optimizes workload distribution, significantly enhancing overall data center efficiency.
- The research was conducted by a team including Gohar Chaudhry, Ankit Bhardwaj, Zhenyuan Ruan, and Adam Belay, and will be presented at the USENIX Symposium on Networked Systems Design and Implementation.
- Future developments may include new protocols for better data placement and leveraging predictability in AI workloads to further enhance SSD operations.
- The research received funding from the National Science Foundation, the U.S. Defense Advanced Research Projects Agency, and the Semiconductor Research Corporation.
data centersworkload optimizationflash storagehardware efficiencyperformance improvement