支援超低延遲儲存裝置之混合程式執行的延遲感知輸出入排程

Abstract

近年來許多服務會執行在邊緣伺服器上，因為其提供了低延遲時間的好處。為了提升系統的效能，我們會使用 Linux 的輸出入排程方法來規劃輸出入順序，並且搭載超低延遲儲存裝置提升資料存取的速度。然而，我們觀察到在混合程式執行時，那些輸出入資料使用量大的程式執行，會影響到對於資料搬移延遲具有需求的程式執行，而導致其無法在規定的服務品質響應時間內完成。因此，我們提出延遲感知的輸出入排程，確保延遲感知程式執行時的服務品質。結果顯示延遲感知輸出入排程策略可以有效地減少錯失服務品質的響應時間，並且不會犧牲過多系統效能。

In recent years, many services have been running on edge servers to take advantage of the low latency performance. To improve system performance and reduce data access time, we employ the I/O scheduler in Linux system to arrange the order of I/O operations with ultra-low latency storage devices. However, we observed that processes with higher I/O data transfer rate would impact the execution of latency-sensitive processes under hybrid task executions, leading to a failure to meet the Quality of Service (QoS) requirements for response time. Hence, we proposed a latency-aware I/O scheduling strategy to ensure the service quality during latency-sensitive process execution. The results show that the latency-aware I/O scheduling strategy effectively reduces the response time of missed Quality of Service (QoS), without significantly sacrificing system performance.