The Effect of Memory-Management Policies on System Reliability

Abstract

Memory management policies are found in many areas of computer systems (eg, virtual memory, cache, database buffers). They have typically been studied in regard to their structure and performance. This work proposes a new study area: the impact of memory-management policies on memory reliability. The tradeoffs in performance and reliability are studied as a function of the block-miss reload time. A detailed analysis of how reliability is affected by the memory space allocated is presented. This can be used to understand the relationship between the amount of memory allocated and reliability. Separate effects are measured, depending on the relative time to process a first level memory miss: Inverse Performance-Reliability Relationship; The reliability improves as the performance decreases. This improvement comes from forcing the blocks out of memory more quickly and essentially scrubbing them. This occurs for small miss times. Dual Reliability-Performance Optimums: The reliability has a maximum value corresponding to a minimum memory usage. This occurs for intermediate miss tunes (eg, a time equal to 104 memory references). Directly Related Performance-Reliability: Improved performance leads to unproved reliability. For large miss times, the residency time is dominated by the miss processing. Since the policy with the smaller memory produces more misses per residency there is a higher duration per residency. For very small memories, a very few page durations contribute to a majority of the total unreliability. Two techniques are suggested to remove these long durations, which lead to even greater improvements in the reliability. 1) An algorithm, selective scrubbing, to break the long durations, which could either be implemented in software or hardware. 2) The addition of very small amounts of duplexed memory can also lead to important reliability improvements. © 1993 IEEE