Improving the efficiency of data centers, the very backbone of the digital world we thrive in, has always represented a formidable challenge. A frequent strategy involves pooling numerous storage devices over a network to allow multiple applications to share resources. However, despite this intelligent approach, a significant amount of capacity often remains underutilized due to the persistent performance variability across different devices.<\/p>\n
Determined to address this issue, a group of astute researchers from MIT managed to create an innovative system that tackles not one but three major sources of variability simultaneously. This leap in technology delivers a considerable boost to the performance of storage devices, far outpacing traditional methods which typically only address one source at a time.<\/p>\n
The core of this transformative system is a two-tier architecture. Task assignments are made by a central controller, while lesser, more immediate issues of data rerouting in case of a struggling device are handled by local controllers. This layout is easily adaptable to real-time changes in workloads and doesn’t require any specialized hardware. In practical tests involving tasks such as AI model training and image compression, this system doubled the performance over traditional approaches, significantly enhancing data center efficiency.<\/p>\n
The driving force behind this research, Gohar Chaudhry, an EECS graduate student and also the lead author of a paper on this fascinating technique<\/a>, stressed on the importance of maximizing the utilization of expensive and environment-intensive resources. His mantra is clear – extract as much performance as you can from your existing devices before considering replacements. “With our adaptive software solution, you can still squeeze a lot of performance out of your existing devices before you need to throw them away and buy new ones,” Chaudhry expounded.<\/p>\n Being a concerted effort, the research team also includes Ankit Bhardwaj, Zhenyuan Ruan, and senior author Adam Belay. Their pioneering work is set to be presented at the prestigious USENIX Symposium on Networked Systems Design and Implementation.<\/p>\n Solid-state drives (SSDs) form the basis for high-speed digital storage. When you pool together multiple SSDs, it allows shared application use, enhancing overall efficiency. But, it’s not all smooth sailing. Different SSDs perform at different speeds, and the slower ones can become a bottleneck, impeding the overall performance of the pool. This performance variability is primarily caused by differences in SSD hardware, the nature of tasks running, and unpredictable garbage collection processes.<\/p>\n The MIT researchers’ solution, termed Sandook (translated to “box” in Urdu), is a software-based system that addresses all these performance-crippling factors all at once. Through global and local scheduling tactics, Sandook optimizes task distribution and reacts swiftly to critical events, respectively. For instance, it shifts operations away from congested devices and minimizes read-write interferences. Sandook also tailors workload based on the individual characteristics and capacity of each SSD.<\/p>\n Moreover, this intelligent system also manages variabilities that occur over different time scales, from unexpected garbage collection delays to wear-induced latency spread over several months. Tests on a pool of ten SSDs revealed that Sandook boosted throughput by 12 to 94 percent compared to static methods while improving SSD capacity utilization by a colossal 23 percent. And it achieved all of this without any need for specialized hardware or updates.<\/p>\n Going forward, the researchers have set their sights on incorporating new protocols on the latest SSDs for better data placement control and leverage the predictability of AI workloads to further enhance SSD operations. Josh Fried, a software engineer at Google, praised their work saying, “Flash storage is a powerful technology underpinning modern data center applications. This work moves the needle meaningfully forward with an elegant and practical solution ready for deployment.”<\/p>\n This revolutionary research was made possible by the funding received from the National Science Foundation, the U.S. Defense Advanced Research Projects Agency, and the Semiconductor Research Corporation.<\/p>\n Intrigued by the power of Artificial Intelligence and seeking its applications for your company? Visit implementi.ai<\/a> to explore how AI can revolutionize your business operations.<\/p>\nHow Does It Work?<\/h5>\n
What’s in Line for the Future?<\/h5>\n