On January 23rd, Intel participated in the SNIA Persistent Memory Summit in California. We have come a long way in the last eight years; we no longer spend most of our time explaining what persistent memory is, but instead talk about the latest products, start-ups, and research ideas going on around it. One thing was clear from the summit—the state of the ecosystem is rapidly maturing, and we are seeing some innovative use cases with Intel® Optane™ persistent memory (PMem).
Intel launched Intel Optane PMem at the Intel Data Centric day in April of last year and many companies were ready with supporting software on our launch date. Since then, we have continued to see our customers and partners drive solutions that solve critical customer challenges. Here are some of my favorite examples from the SNIA Persistent Memory Summit.
One of the high points of the day was Oracle’s presentation by Jia Shi. She walked through why just treating persistent memory as “faster flash” would leave a bunch of performance on the table. So instead of doing that, she showed two ways that Oracle Exadata was modified to fully leverage persistent memory with very impressive results. First, she demonstrated using Accelerated RDMA to read directly from a database cache that Oracle built using persistent memory. Second, she showed using Accelerated RDMA to write directly to Exadata’s database log, which is placed in persistent memory on the Exadata server. The best part was that everything she talked about has been shipping as part of every Exadata system since September, and all these systems contain 1.5 TB of Intel Optane PMem.
Yao Yue from Twitter presented on enabling Pelikan for Intel Optane PMem. Pelikan is a cache framework created as part of Twitter's unified cache project that provides an expanding collection of cache services, and a common library used to build them. Yao provided a brief overview of the Pelikan architecture and how it was modified to leverage persistent memory. Yao shared that with (less invasive) modest code changes and a number of Intel developer hours, Twitter sees server performance comparable to DRAM for tested Twitter workloads and excellent recovery performance.
Yao shared a line that was repeated later during the summary session: “Memory Mode is a gateway drug." What she meant by this is that Memory Mode is so easy to use that you can quickly see some value from persistent memory, but then she shared performance results showing how much more value she was able to measure using App Direct mode on a modified version of Pelikan.
For more information on Yao’s Talk, check out her presentation.
Intel also participated with some talks of our own—Andy Rudoff shared an overview on the latest activities around persistent memory programming, including within SNIA, academia, and across the ecosystem. Piotr Balcer discussed the evolution of the Persistent Memory Development Kit (PMDK), and shared the lessons he learned during its development along with usage by applications. There were some good questions from the audience, reflecting the serious persistent memory experience in the room. Milind Mittal and Scott Shadley stepped in at the last minute and presented on Compute Express Link (CXL) combined with persistent memory. The talk was very informative and from the questions in the audience, you could tell that people are starting to think about the possibilities when you combine persistent memory on CXL.
Dave Eggleston shared a summary of the different persistent memory technologies—providing the audience with a step back to see the large variety of activity around persistent memory taking place in the industry.
It is an exciting time for persistent memory and the SNIA Persistent Memory Summit was a great way to kick off 2020! I am looking forward to sharing more about the work we are doing in the industry to solve critical business problems with Intel Optane PMem.
You can find a list of all the presentations here: https://www.snia.org/pm-summit
Check out intel.com/optane for more information on persistent memory.