Penumbra: Persistence across the Memory Bus


  • The last decade has seen many breakthroughs in byte-addressable persistent memory (PM) technologies such as phase change memory (PCM), magnetoresistive RAMs (MRAMs), memristors, and various forms of persistent hybrid memories such as NV-DIMMs. The consensus is that these technologies will provide the byte-addressability of DRAM with performance close to that of DRAM, while providing high-capacity, non-volatile storage. These technologies are poised to make their way into storage systems in a few years. Some, such as NV-DIMMs, already have begun market penetration.

    The question is: how can software systems, enterprise systems in particular, leverage these technologies to drastically improve application performance? The Penumbra (Persistence across the Memory Bus) project is centered around answering this question. We are interested in building new system infrastructures, and rearchitecting existing ones, with the perspective of exploiting PMs for next generation enterprise systems.


I joined Oracle in September 1988 to take over responsibility for Oracle V6 buffer cache and recovery. I implemented the first version of Oracle Parallel Server. I redesigned and rewrote much of the recovery code for Oracle 7 and 8i. I contributed to the design of cache fusion for RAC. I specified requirements for storage to support Oracle. I came up with the idea for ASM, and lead the ASM design effort starting in 2000. I have been working with Sun engineers to better integrate their storage products with Oracle. Lately I have become interested in how Oracle can use persistent memory.


1. Faisal Zaghloul (Yale University) summer of 2016.
2. Yihe Huang (Harvard University) summer of 2016.
3. Sanidhya Kashyap (Georgia Institute of Technology) summer of 2016.
4. Amee Trivedi (University of Massachusetts Amherst) summer of 2015.
5. Achin Mishra (University of Wisconsin-Madison) summer of 2015.
6. Nick Murphy (Harvard University) spring and summer of 2014.