Execution of a Soft Real-Time Multitask Program on NAND Flash Memory with Guaranteed Reliability

Abstract

As solid-state drives based on flash technology are becoming a staple for persistent data storage in data centres. Especially, solid-state drives based on the NAND flash technology in non-volatile storage for storing data. Yet it is still facing some difficulties in executing program codes on a NAND flash memory as a consequence of supporting only page-based reads. But with the constructed method of recent years so-called mRT-PLRU (Multitasking Real-Time constrained combination of Pinning and LRU), which forms a generic framework to use inexpensive non-volatile a NAND flash memory storage for storing and executing real-time programs in a multitasking environment, it allows executing program code on a NAND flash memory. However, since this technique does not consider reliability problems it exhibits some issues when performing read operation - especially read disturb errors which occur after repetitive read operations for each block, there is a limitation in terms of reliability in the execution of a real-time program code. Therefore, we propose a new approach to execute soft real-time tasks on NAND flash memory with minimal RAM size while considering read disturb errors. For this, our approach has been performed in two steps in order to execute multiple real-time tasks stored in a NAND flash memory with the minimal requirement of a RAM size. In the first step so-called, per-task analysis, we analyse execution time of each task one by one by changing RAM size and check blocks in NAND flash memory which are accessed during execution of the task corresponding to optimal pinning and LRU combination for each individual task. In the second step so-called, stochastic-analysis-in-loop-optimization, we use those functions from the per-task analysis for all tasks as an input to allocate minimum possible RAM size to execute all tasks while meeting given deadline meet probabilities. In this step, we consider additional delay needed to prevent read disturb errors. As a consequence of configured technique, compared with other approaches, this technique significantly minimised a RAM size requirement. Motivated by this, under this kind of real-time system circumstance, we propose a new aproach to guaranteeing reliability, especially focusing on the read disturb problem on the NAND flash memory while executing multiple tasks. To this end, our approach rewrites the blocks considering overhead for a rewrite of NAND blocks. Those blocks with high read count, before exceeds the threshold limit for each block, should be rewritten earlier with the reserved time slots. At the end, checking of all given tasks schedulability with an exact stochastic analysis for probabilistically guaranteeing the reliability of read operation on NAND flash memory that the deadline is met for all given tasks.