This Software boosts Data Centre performance efficiency by optimizing storage workloads, unlocking hidden speed without adding costly hardware.
Improving data centre performance has often meant adding more hardware, but that approach is becoming increasingly costly and unsustainable. Researchers have now developed a smart software-driven solution that significantly enhances storage performance without requiring additional physical resources.
Modern data centers commonly pool multiple solid-state drives (SSDs) so that applications can share storage capacity efficiently. However, this setup often leaves a considerable portion of performance untapped due to variability among devices. Differences in hardware age, workload patterns, and internal processes can cause some drives to lag, ultimately limiting the performance of the entire system.
To address this challenge, researchers at MIT Computer Science and Artificial Intelligence Laboratory introduced a system called Sandook, designed to intelligently balance workloads across SSDs. The system tackles three major sources of performance variability simultaneously: hardware differences between drives, interference between read and write operations, and unpredictable slowdowns caused by garbage collection processes within SSDs.
Sandook operates using a two-tier architecture. A global controller oversees the entire storage pool, assigning tasks based on each device’s capabilities and condition. Meanwhile, local controllers on individual machines respond instantly to performance fluctuations, redistributing workloads if a device begins to struggle.
One key innovation that reduces the conflicts slowing down performance is how Sandook separates read and write operations across different drives. It also continuously profiles each SSD, allowing it to detect when internal processes like garbage collection are likely to cause delays. When this happens, the system temporarily reduces the workload on the affected device and shifts tasks elsewhere, maintaining overall efficiency.
In testing, Sandook demonstrated impressive results across multiple real-world scenarios, including database operations, AI model training, and image compression. It improved application throughput by up to 94 percent and increased overall storage utilization by 23 percent. Notably, the system enabled SSDs to reach up to 95 percent of their theoretical maximum performance.
This approach highlights a shift toward smarter resource management in data centers. By maximizing the potential of existing hardware, solutions like Sandook can reduce costs, extend device lifespan, and lower environmental impact, without compromising performance.
