Systems software support for new generation non-volatile memories

点击翻译按钮获取中文摘要

This NSERC Discovery Grant proposes novel and timely research on systems software support for the new, emerging non-volatile memory (NVRAM) technologies. These new technologies operate at byte granularity, are almost as fast as DRAM, and yet they have a price point one order of magnitude lower than DRAM. Two examples of potential NVRAM technologies are phase-change and memristor memories. ******Disruptive nature of NVRAM technology: We believe these new NVRAM technologies will be game changers, and this prospect has us very excited. NVRAM will be disruptive to both computer systems architectures and to applications. NVRAM will disrupt computer systems architectures because it will either require a new layer in the memory hierarchy or it will partially merge and replace the DRAM and local, persistent storage layers that have permeated computer systems for the last 50 years. NVRAM will disrupt applications because accessing persistent storage will be dramatically more efficient than with current memory technologies. We even anticipate new programming models.******Effect on systems software: These new realities will affect systems software, the layer of software between hardware and applications, in a major way. Existing systems software will have to be adapted to the new hardware architectures. For example, file systems and virtual memory systems will have to be completely redesigned; the organization of the operating system and its servers will likely have to be completely reorganized; and the model of protection will likely have to be significantly altered. Furthermore, new systems software will need to be devised to support applications that use the new memories and to support the anticipated new programming models. ******NVRAM offers many compelling potential advantages, but at the same time introduces challenges. For example, long-running applications can be made much more reliable by keeping all application states in NVRAM, thus allowing them to continue execution after many types of system failures. On the other hand, system failures can result in permanent inconsistencies if appropriate interventions are not taken.******Research objectives: Our objectives for this research are: (i) to obtain a better understanding of how best to use NVRAM so as to benefit applications and systems software, (ii) to develop general techniques to help applications better utilize NVRAM, (iii) to advance the state-of-the-art in systems software to accommodate NVRAM technologies, (iv) to enable new classes of applications that make use of NVRAM, and (v) to provide a fertile training ground for HQPs.**