Researchers Develop New Method to Boost SSD Speeds

Over the past 2 years, the SSD industry has been undergoing a transformation to remove the interface bottleneck. Moving past legacy ACHI in SATA3, NVMe is now the gold industry standard for SSDs. Designed from the ground up for flash storage, the new interface allows SSDs to stretch their legs. Research is now turning to other bottlenecks exposed by NVMe now that the interface is no longer holding things back.

Researchers at the Yonsei University in Korea have come up with a new way to schedule flash memory accesses. As a result of their research, they have come up with a new I/O (input/output) scheduler. Dubbed Dynamic Load Balanced Queuing (DLBQ), the scheduler tries to distribute access requests evenly across memory chips to avoid contention. By reducing the access to each NAND die at a time, there is less contention for resources.

DLBQ Offers 11-15% Speed Increase for SSDs

One of the big bottlenecks SSDs have run into is the number of NAND dies accessible. Each SSD is made up of several NAND dies. Right now, SSD controllers generally try to spread out data writes even between dies for wear levelling purposes. However, DLBQ applies more widely to all accesses to spread them out evenly between dies. By holding back certain requests, each access is given full speed and less overhead spent on juggling multiple requests at the same time. In addition, DLBQ is latency sensitive so requests that are held back have higher priority.

Currently, the main way SSD capacities are increasing is through denser dies. For smaller SSD sizes, this means fewer dies to spread the workload. Since each die has limited bandwidth and resources, smaller SSDs tend to suffer under heavy load. With DLBQ offering up to 15% speed improvement, it may make its way into drives soon. This will hopefully allow smaller drives to exist as viable storage devices as NAND die sizes continue to climb.