COMeT: Continuous Online Memory Test

Abstract

Abstract-Today's computers have gigabytes of main memory due to improved DRAM density. As density increases, smaller bit cells become more susceptible to errors. With an increase in error susceptibility, the need for memory resiliency also increases. Self-testing of memory health can proactively check for errors to improve resiliency. This paper describes a software-only selftest to continuously test memory. We present the challenges and design for an approach, called Continuous Online Memory Testing (COMeT), that targets chip multiprocessors. COMeT tests memory health simultaneously with application execution in anticipation of allocation requests. The approach guarantees that memory is tested within a fixed time interval to limit exposure to lurking errors. We developed and evaluated an implementation of COMeT. On the SPEC CPU2006 benchmarks, COMeT has a low 4% average performance overhead. When emulated errors were injected into physical memory, applications executed 1.13x to 4.41x longer with COMeT than without it.

Asteroid

“…The diagnostic can be implemented in software, such as Memtest86+ [1]. In an online setting, assurance guarantees can be made that physical memory has been tested within some fixed time period, i.e., a vulnerability window [11,12,13].…”

“…As motivation and background for Asteroid, we describe a previous online technique, COMeT+ [11,12]. This technique tests memory online and assures the vulnerability window for multi-core machines with low core counts (e.g., dual or quad core).…”

“…After the whole page is written, the page is read again to check that the patterns were written correctly to DRAM. Burst writes dramatically reduce test latency compared to writing one word at a time bypassing the caches [11]. Test threads also implement strided writes and reads to flush DRAM row buffers before reading them back.…”

“…Recent research demonstrated the benefit of an online memory diagnostic to guarantee that memory used by an application workload has been tested prior to allocation [11,13,12]. A single diagnostic thread runs in the background on an idle CPU core to test free physical pages prior to allocation [12].…”

RDIS: A recursively defined invertible set scheme to tolerate multiple stuck-at faults in resistive memory

“…While it is an issue orthogonal to the goal and scope of this work, such retirementbased bad memory block management is commonly used in storage devices [32], [33] and is becoming increasingly important for main memory as well [34], [35].…”