Abstract
The advancement in I/O technology has posed an unprecedented demand for high-performance processing on I/O data, leading to the development of Data Direct I/O (DDIO) technology. DDIO improves I/O processing efficiency by directly injecting all inbound I/O data into the last-level cache (LLC) in cooperation with any type of I/O device. Nonetheless, it has been observed that DDIO can cause significant interference between I/O applications inside the LLC, resulting in the degradation of system performance. Especially, in this paper, we demonstrate that storage I/O on modern high-performance NVMe SSDs hardly benefits from DDIO, sometimes causing inefficient use of the shared LLC due to the “leaky DMA problem”. To address this problem, we propose LADIO, an adaptive approach that mitigates inter-application interference by dynamically controlling the DDIO functionality and reallocating LLC ways based on the leakage and locality of storage I/O data, respectively. In scenarios with heavy I/O interference, LADIO improves the throughput of network-intensive applications by 20% while maintaining that of storage-intensive applications.
Ipoom Jeong
Assistant Professor
My research interests include CPU/GPU microarchitectures, memory/storage system designs, and smart-I/O devices