Persistent Memory Summit 2020 Presentation Abstracts

2019 Persistent Memory Summit Presentations


Keynote: The Evolution of Cloud-scale Storage from iSCSI to RoCE to NVMexpress over Ethernet to NVMexpress over TCP/IP

Andy Bechtolsheim, Chief Development Officer, Co-Founder, and Chairman, Arista Networks

Abstract

Mr. Bechtolsheim will present to a joint audience of SNIA Annual Members Symposium and 2020 Persistent Memory Summit attendees.  His keynote session will discuss the evolution of cloud-scale storage from iSCSI to RoCE to NVMexpress over Ethernet to NVMexpress over TCP/IP.


The Persistent Memory Programming Model

Andy Rudoff, Member, SNIA NVM Programming Technical Work Group and Persistent Memory SW Architect, Intel Corporation

Abstract

In the seven years since we set out to define the programming model for persistent memory, we’ve seen a flurry of activity around OS enabling, application support, and some interesting new use cases. This session will summarize the latest activities around persistent memory programming, both inside SNIA, in academia, and across the ecosystem.


PM in the Wild: VMware Experience and Future Expectations

Richard A. Brunner, Principal Engineer, CTO Server Platform Technologies, Vmware

Abstract

Since mid-2017, VMware has been supporting various forms of PM in our hypervisor products. It has been a rough honeymoon.

This session discusses VMware's POV, experiences, and challenges with PM so far. We conclude with a brief survey of promising trends that will improve the situation.

Back to Top


Persistent Memory use cases in modern software architectures

Soji Denloye, Software Development Engineer, Intel Corporation

Abstract

There are many applications that can benefit from persistent memory and, thanks to its versatility, many ways to bring about these benefits. We are learning that, for a given application, the best solution involves choosing how to apply persistent memory: as buffers, caches or storage and either in a persistent or volatile way. This talk will describe scenarios where persistent memory enabling of frameworks such as Cassandra, MemcacheD and RocksDB resulted in improved performance.


Memory at Storage Scale, Storage at Memory Speed

Charles Fan, Co-Founder and CEO, MemVerge

Abstract

Future data centers will be memory-centric and software-composed.  Storage Class Memory (SCM) is a key enabling technology to this future.  SCM+DRAM, combined with the right software, become the new Memory Converged Infrastructure (MCI) layer that is larger in capacity than the memory layer today, and persistent.  It can change the way applications are developed and deployed, and make primary storage unnecessary.  In this talk we'll introduce MemVerge MCI software technology and its initial use cases.

 

Evolution of the Persistent Memory Development Kit

Piotr Balcer, PMDK Architect, Intel Corporation

Abstract

Starting from a couple academic papers and no actual persistent memory, the PMDK team made its first commit in 2014 and began a long journey of invention and learning. Persistent memory programming can be simple, given the right high-level APIs, and it can be complex and error prone when more flexible and powerful APIs are used. Meeting application requirements and constructing useful APIs requires years of experimentation, learning from mistakes, and integrating feedback from a large ISV community. Lead PMDK architect Piotr Balcer will talk through some of the lessons learned during PMDK development and usage by applications.


The Perfect Trifecta. NVMe, CXL and Persistent Memory!

Millind Mittal, Fellow, Xilinx and Scott Shadley, VP Marketing, NGD Systems

Abstract

The Compute eXpress Link (CXL) has the potential to bring coherency (sic) to the world of open, off-chip coherent busses. As such it will become an innovation point for the deployment of Persistent Memory and acceleration. In addition, thanks to its backward compatibility, CXL also has support for devices which include NVMe capabilities for legacy block and key-value storage, memory centric capabilities like CMB and network-centric capabilities like NVMe-oF. In this talk we will introduce the audience to CXL and how it can be used as an innovation platform to extend existing NVMe devices to the new and exciting world of CXL and Persistent Memory. We will discuss the nuts and bolts of CXL, the upgrade path from NVMe to CXL, a variety of use-cases for NVMe/CXL devices and software issues like drivers, OS support and management.


Persistent Memory: Media, Attachment, and Usage

Dave Eggleston, Principal of Intuitive Cognition Consulting

Abstract

The commercialization of Persistent Memory is well underway, fundamentally changing the way processors load and store information. The characteristics of the non-volatile memory media (3DXP, PCM, MRAM, RRAM, others) determine the basic performance, reliability, and cost that Persistent Memory can deliver to the system. Brand new methods of processor attachment (DDR5, DDR-T, NVDIMM, CXL, CCIX, Gen-Z, OpenCAPI) for Persistent Memory promise to broaden its usage far beyond “DRAM replacement”, opening up vast resources enabling memory and power hungry applications.

In this talk, the speaker will:

  • Discuss the basic characteristics of Persistent Memory Media
  • Articulate how Persistent Memory enhances server architectures
  • Identify, compare, contrast the new methods for Persistent Memory attachment
  • Propose innovative ways applications will utilize Persistent Memory to overcome memory and power limitations
  • Discuss who is doing what within the Persistent Memory ecosystem

 


Providing Native Support for Byte-Addressable Persistent Memory in Golang

Pratap Subrahmanyam, Fellow, VMware

Abstract

Persistent memory provides various benefits. First, its high density allows for large capacity random access storage to be situated close to compute. Second, its memory like latency allows for applications to directly place data structures that are latency sensitive in them. We fully expect the first value proposition to be the leading factor in the adopiton of persistent memory, primarily because it requires no changes to the applications. But, for full adoption of persistent memory, we need to make programming for byte addressable persistence easy. There are several efforts in this direction, and I will describe one of them that we are conducting in VMware, where we have ported the Golang compiler to have native support for persistent memory. We then use this compiler to develop a Redis compatible key value store, and we show that our compiler modifications make it easier to touch and feel byte addressable persistent memory from an application.

Back to Top


Oracle's Unique Implementation of Persistent Memory for Accelerating Database Workloads

Jia Shi, Sr. Director of Software Development for Exadata, Oracle Corporation

Abstract

Persistent memory is a new silicon technology, adding a distinct storage tier of performance, capacity, and price between DRAM and Flash. The persistent memory is physically present on the memory bus of the storage server resulting in reads at memory speed, much faster than flash. Writes are persistent, surviving power cycles, unlike DRAM. Oracle has engineered Exadata Smart PMEM Cache and Exadata Smart PMEM Log functions to achieve this significant boost in Oracle Database performance.

Jia Shi, Sr. Dir of SW Development will describe the engineering details of their implementation in Exadata and how Oracle's software development teams are innovating with this technology.


Using Persistent Memory with Pelikan

Yao Yue, Software Architect, Twitter, Inc.

Abstract

Pelikan is an open sourced, module cache developed by Twitter.  It is designed for high performance and high reliability.  In this session, Yao Yue will provide a brief overview of the Pelikan architecture and how it was modified to leverage persistent memory.


Introduction to PM Hackathons

Jim Fister, Principal, The Decision Place; Director, SNIA Persistent Memory Software Enabling

Abstract

At the 2020 PM Summit, we've seen examples of successful companies developing for persistent memory in the application base. It's obvious that there's benefit to be found in persistent memory. SNIA and its participant companies have put a focus on creating an environment where it's possible for any programmer to learn the basics of software development using the PMDK toolkit. This session will provide an overview of the options companies have for educating the developer community on programming PM. It will also highlight educational tools and events that the company can utilize. Whether it's educating a few individuals, or the entire company's development organization, SNIA is ready to help our partners move to the new development paradigm.


Using Real World Workloads and Artificial Intelligence to Optimize NVMe SSD and Persistent Memory Performance

Eden Kim, CEO, Calypso Systems; Yafei Yang, CEO, DapuStor

Abstract

In this session, see how we evaluate NVMe SSDs (including low-latency NVMe SSDs) and Persistent Memory performance on real world workloads.  We also use real world application workload captures as training for AI optimization with Long Short Term Memory (LSTM) Recurrent Neural Network (RNN) Artificial Intelligence to optimize storage performance in NVMe SSD.  Calypso IOProfiler real world application captures are used to optimize workload IO Stream recognition and IO determinism in storage applications and AI optimized workloads.  DapuStor AI Machine Learning (ML) training and LSTM RNN technologies are used to improve SSD wear leveling, task scheduling, pre-fetching and caching, and to minimize write amplification and garbage collection.  We compare and analyze different performance data in Optane Persistent Memory (DCPMM), NVDIMM-N, XL Flash based NVMe SSD, Optane 3D XPoint NVMe SSD and 3D NAND NVMe SSD on a variety of real world application workloads.