Here are the Flash Memory Summit 2020 sessions that took place on the first day of the virtual conference. Our content library is split into sections to help you get the most of your experience and maximize your time. Follow the links to view the keynote presentations, content from the second day and content from the closing day.
The sessions from the first day of the Flash Memory Summit event span six tracks, including: special presentations, hyperscale applications, NVMe and NVMe-oF, enterprise systems, new memory technologies and software.
The content featured in the special presentation section include insights from TechTarget's Vice President of Market and Digital Strategy Jillian Coffin and Executive News Director Dave Raffo, as well as the Lifetime Achievement Award ceremony and more.
Whether you are looking to better understand the challenges of deploying in hyperscale, need a refresh on the latest NVMe technologies or want insights to help further your skills, check out the sessions that happened on day one and take a deeper dive into the presentations that really stick with you.
Session C-1: Advances in Enterprise Flash Storage
This session is divided into two parts: Part 1 is an annual update on enterprise storage and part 2 is a panel on data growth and a future with zettabytes.
Annual Update on Enterprise Storage
Enterprise flash storage continues to advance to being a storage layer on its own. Products based on these technologies are currently being deployed in enterprises and their numbers will rapidly increase in the near future. Link to Presentation
Managing Data Growth in the Zettabyte Era (Panel)
Data centers must handle more data all the time. Terabytes have given way to petabytes and now to zettabytes. AI/machine learning, IoT and analytics are all affecting the explosive growth of data and how it impacts storage capacity and performance. Everyone involved in managing or analyzing data must look for new ways to provide storage with higher capacity and faster performance. Join our panelists who discuss these topics. Link to Presentation
Session D-1: Smart New Architecture for Cloud-Scale Storage
Smart New Architecture for Cloud-Scale Storage
Today's huge and ever-growing data stores require new architectures to provide the required capacity, performance and flexibility. Such architectures must be based on the latest technologies such as 3D XPoint, QLC flash and NVMe-oF. They must also include extensive software that provides the required management tools and utilities, as well as specific algorithms to handle the new hardware.
Session C-2: Top Ten Things You Need to Know about Enterprise Storage Today (Panel)
Top Ten Things You Need to Know About Enterprise Storage Today
Flash memory has morphed rapidly from an exciting new technology that had to be justified for specific use cases to a standard part of every data center. What other kind of storage would you be buying today? What do vendors and users alike need to know about flash? Is it the emergence of 3D technology, the rapid rise of the high-speed NVMe standard, the promise of persistent memory (storage at memory speeds), the role of flash in scalable systems (such as clouds and megawebsites), new methods for flash storage networking (such as NVMe-oF), ways to make software take advantage of flash memory, or large, hierarchical storage systems that cover everything from high-speed cache to long-term archiving? Our top industry experts present their own candidates for the top ten list. We then open nominations to the audience and finish with our vote for the top ten for 2020. Link to Presentation
Session A-1: What the Hyperscalers Are Buying
About 50% of all storage sold worldwide goes to hyperscale applications, so getting hyperscalers the right storage at the right price has become a major industry concern. Hyperscale storage must obviously be large, fast, highly scalable, easy to administer, low-latency, low-cost, and able to handle a wide (and largely unknown) variety of applications. Emerging technologies of great interest to hyperscalers include low-cost archival drives (using QLC technology), persistent memory, NVMe, NVMe-oF, zoned namespaces and Ethernet-attached drives. With the huge size of hyperscale data centers comes an increasing emphasis on unit price, automated management, high reliability and easy identification and removal of failing parts.
Challenges in Hyperscale: What Hyperscalers Care About
This talk focuses on the challenges of deploying in hyperscale. What the problems and industry needs are. What hyperscaler's care and do not care about. Link to Presentation
SSD Reliability and Debug at Scale
There are some really exciting innovations in the world of storage. For physical memory we have QLC, Optane and ZNAN and from the specification side NVMe has added IO determinism, sets and now Zoned Namespaces. With all these advancements we still have a high amount of SSD failures in production. These failures bring with them unique challenges for debugging them and designing them to be reliable. This presentation describes some of the challenges around SSD debugging that we face at scale at Facebook and how we overcome these challenges. The session suggests how the storage industry can come together and help advance this area. Link to Presentation
Using Persistent Memory to Accelerate Tweet Search
Twitter's in-house distributed tweet search services are built around the Apache Lucene text search engine. As an option to improve per-node throughput we sought to test the storage of tweet data in persistent memory instead of SSDs. Persistent memory offers lower access latency compared to SSDs and allows mmap to establish direct mappings, bypassing the page cache. Because our service is primarily memory-bound this was an appealing feature. The trade-off is less storage than with SSDs. Link to Presentation
Handling Slow Disks in Heterogeneous SSD Deployments
In large distributed system deployments, it is common to have a heterogeneous fleet of solid-state storage devices. Such fleets introduce challenges for latency-sensitive applications such as content delivery. As wear levels increase, different SSD models show different latency profiles. Although some might maintain a consistent average latency, they would show a higher variance, which would have an impact at the application level. In this talk, we show the performance differences between a few SSDs models. We also provide insights on how we address this problem using statistical models. Link to Presentation
Session A-2: Hyperscale Storage in 2025 and How We Got There (Panel)
Hyperscale Storage in 2025 and How We Got There
Hyperscale storage in 2025 is likely to swallow up an ever-increasing percentage of the overall storage market, as organizations farm out more of their IT and focus more of their energy on their core competencies. Hyperscalers will require ever more complex storage systems with all storage being networked for maximum utilization, banks of low-cost flash systems used for archiving of items needed only occasionally, and multi-tiered, automated main storage systems with small amounts of new technologies used for persistent memory and high-speed caching. Link to Presentation
NVMe and NVMe-oF
Session B-1: NVMe-oF: The Best Way to Network Enterprise Storage
The latest NVMe over Fabrics specification (Version 1.1) is now available to accompany the NVMe 1.4 specification. The combination introduces important new functionality such as the NVMe/TCP transport and ANA (asymmetric namespace access) multipathing. NVMe-oF is quickly gaining traction as a network technology enabling new architectures and use cases for disaggregated and hyperconverged cloud deployments. What does it take to deploy NVMe-oF technology in data centers? Practical experience is now available to suggest ways to raise performance levels for NVMe/TCP networks to meet the demands of a wide variety of applications.
Storage Environments with NVMe-oF Technology
In this presentation, explore what NVMe, NVMe-oF and what it enables, and how Kubernetes works. Learn how the NVMe-oF protocol enables a new deployment of storage called desegregation that enhances Kubernetes deployment. Link to Presentation
Bringing NVMe/TCP Up to Speed
In this presentation, get an overview of NVMe over TCP and learn about the latest improvements to improve NVMe-TCP performance. Link to Presentation
Accelerating Performance with NVMe-oF
Can you accelerate your performance with NVMe-oF? Join George Crump, former storage industry analyst and CMO at StorONE, to learn if NVMe-oF is ready for your data center and if you are ready for NVMe-oF. In his presentation, George compares the pros and cons of NVMe-oF and provides valuable insight for enterprise contemplating the new protocol. Attend to learn if you can really accelerate performance with NVMe-oF. Link to Presentation
High-Performance RoCE/TCP Solutions for End-to-End NVMe-oF Communication
Exploiting the full SSD performance in scalable disaggregated architectures is a continuous challenge. NVMe/TCP, released in 2018, enables a broader sharing of distributed storage resources. It complements NVME-oF over RDMA, avoiding performance degradation over distant links and simplifying the deployment. However, this comes at the cost of an heaviest networking stack and requires the latest Linux kernels. In this talk, we analyze the differences between RoCE and TCP and show how to eliminate bottlenecks and achieving best-in-class performance for both protocols in an end-to-end NVMe-oF communication. We also demonstrate how this solution can be OS agnostic, ensuring a seamless integration of NVMe-oF in today's data center.
Session B-3: Getting the Most Out of NVMe
NVMe is has rapidly become the standard way to achieve high-performance storage with widespread adoption in client, cloud and enterprise applications. It is now widely used in both direct-attached and fabric-attached applications. The NVM Express organization released the 1.4 base specification in 2019, and since then implementers have rapidly increased their utilization of its many new features. Specific examples include Sanitize enhancements, Enhanced Command Retry, Read Recovery Levels, Endurance Groups and Persistent Memory Region. The future focuses on a transition to a merged base specification including both NVMe and NVMe-oF architectures. It will present areas of innovation that preserve the simple, fast, scalable paradigm while extending the broad appeal of NVMe. These continued innovations will ready the NVMe ecosystem for yet another period of growth and expansion.
The State of NVMe Interoperability
VNVMe has become synonymous with high-performance storage with widespread adoption in client, cloud and enterprise applications. Although initially developed for direct-attached PCIe SSDs, NVMe is now widely used in both direct-attached and fabric-attached applications. This presentation provides an overview of the latest NVMe technologies, summarizes the NVMe standards roadmap and describes new NVMe standardization initiatives. Link to Presentation
Preview of NVMe 2.0: Impacts for the Data Center
NVMe architecture has rapidly evolved from a disruptive technology to a core element in storage architectures, unifying client, hyperscale and enterprise applications into a common framework. NVMe version 2.0 specification, the next major upgrade of the specification, will focus on a transition to a specification merger covering both NVMe and NVMe-oF technologies. It will make fabrics a core concept in NVMe specification, eliminate duplication in data structures and integrate NVMe and NVMe-oF base functions. It will create an extensible structure including a modular transport mapping layer that covers PCIe as well as other choices. NVMe version 2.0 will also include other new features, changes and upgrades. The aim is to preserve the original paradigm of simple, fast and scalable, while extending the appeal of NVMe technology to a wider range of applications. The continuing innovations will prepare the NVMe technology ecosystem for another period of growth and expansion. Link to Presentation
NVMe Technology in the Real World
This presentation discusses the real-world capabilities of the NVMe 1.4 and NVMe-oF 1.1 specifications. Attendees will learn about the applications Facebook's hyperscale data center uses, and how they benefit from NVMe features in deploying at scale. Attendees will also learn from HPE about how NVMe over Fabrics is evolving to redefine next generation SAN and the key fabric requirements to enable this new frontier. Link to Presentation
NVMe Market Research
NVMe and NVMe-oF have progressed from the frontiers of technology to everyday use in data centers, clouds, and high-performance computing over the past five years. New features have been added, and software support is now available in all major operating systems. Much more lies ahead. Data centers can expect a convergence of NVMe and NVMe-oF command sets and new features that add more flexibility, simplify management, and allow for use in a wider variety of applications. Link to Presentation
Session B-4: NVMe in Cloud Applications (Panel)
NVMe in Cloud Applications
Large installations such as Facebook, Mellanox, Western Digital and Microsoft use NVMe technology to advance their cloud applications. They have a wide variety of application requirements and challenges, but all of them found good reasons to choose NVMe technology for their cloud storage. Link to Presentation
New Memory Technologies
Session A-3: Using 3D XPoint to Overcome Storage Roadblocks
The speculation surrounding Intel/Micron 3D XPoint technology has been huge, ever since its introduction in 2015. What is the reality? Where is the technology today and what applications are already using it? What are the effects of the relatively high prices (five times that of flash) that have now been suggested? Where is the technology heading and what steps will occur along the way? What are the major use cases that will drive large markets?
Using 3D XPoint Persistent Memory Effectively
There's a lot of excitement around 3D XPoint and Optane persistent memory (PMEM). It falls between DRAM and flash NVMe SSDs. It's five times faster than the fastest flash SSD while ten times slower than DRAM. Versus DRAM, it's non-volatile with higher density and lower cost. Analytics, OLTP, AI and deep machine learning vendors all salivate at the prospect. However, several hidden PMEM potholes derail the most promising efforts, and which frustrate architects of all kinds. Attendees at this presentation will learn how to design around those potholes to gain the value of 3D XPoint. Link to Presentation
Media Aware Storage Engine for 3DX and 3D NAND QLC
3D XPoint (Intel Optane technology) and 3D QLC NAND SSD are two advance technologies. One fills the performance gap between DRAM and NAND-based SSDs while the other reduces the cost gap between SSDs and hard drives. To maximize utilization of two technologies, it requires a media awareness storage engine. This lecture will introduce a standalone key-value/object storage engine called smart key-value data store (SKVDS) that recognizes and optimizes performance on different storage media, such as Intel Optane SSDs and 3D QLC SSDs. This solution is independent from upper applications/services so it can be scaled for wide use cases. This lecture will also present test results for popular key-value solutions such as RocksDB. We also show different SKVDS configuration matrix for different performance, QoS, endurance and cost trending.
Tiering Across Intel Optane and NAND Flash Media
Intel Optane media provides data centers with the new ability to increase endurance while simultaneously reducing costs and boosting performance. Highlighting the endurance value of Intel Optane media, this talk is organized into three parts: First, the mathematically precise theory about conditions under which a data center workload benefits from deploying both Intel Optane SSDs and Intel 3D NAND SSDs to improve endurance and performance at reduced cost; second, three proof points with empirical data showing how our theory applies to popular data center workloads; and third, a generic tiering algorithm that maximizes the endurance and performance at optimal cost for data center workloads.
3D XPoint Optane Memory Markets: Bits, Revenue, Costs
Intel is selling its version of 3D XPoint (called Optane memory media) into applications such as fast SSDs and DIMM memory modules and is working on both the second and third generations of the technology. Micron is in early development and sampling on its technology. Products, technology, competition and costs are shown as well as a detailed update on wafer capacity, bit growth, revenue and new technologies. We also show potential strategies for Micron and Intel to ramp this phase-change memory technology from both a fab supply and end market demand analysis. Link to Presentation
Session C-3: How Data Centers Can Profit From New Memory Technologies
The major new memory technologies have now been around for over 20 years. Acceptance of them has been slow due to a variety of issues. MRAM has found uses in IoT, storage systems and mil/aero applications, while RRAM has recently received much interest in areas such as neuromorphic computing. Market analysts predict a bright future for both with markets in the billions of dollars.
Annual Update on New Memory Technologies
We show all reported memory technologies on the product lifecycle timeline that we have published with technologies from pathfinding single bit research to technologies that are selling in systems today. Each technology is presented with next milestone requirements and time-to-revenue production as well as a risk assessment and modeled success probability. Nanotube, ferroelectric, interface and filament ReRAM, phase change and MRAM are all covered. Link to Presentation
MRAM - Latest High-Performance Applications
This presentation provides a walkthrough of a high-performance application that is utilizing MRAM. Link to Presentation
RRAM - Now Used in IoT, AI, and Data Center Storage
Explore the features and latest developments of resistive RAM memory and additional emerging memories, the advantages of resistive RAM, and the operation and construction of resistive RAM memory devices. Link to Presentation
Session A-4: 3D XPoint in 2025, and How We Got There (Panel)
3D XPoint in 2025, and How We Got There
3D XPoint has emerged as a flash competitor that has actual products, use cases, and support. Initial interest is in extremely high-performance SSDs, caches and accelerators. Manufacturers expect to provide much improved devices soon with higher performance, lower latency, longer lifetimes and lower prices. The ecosystem will expand to include more compatible processors, motherboards, buses and software. Analysts expect markets for 3D XPoint to rise steadily. Link to Presentation
Session C-4: What Will Replace Flash and When?
What Will Replace Flash and When?
Is there life after flash? Or will flash memory keep improving and dominate all NVM technology into the next decade? The panelists peer into their crystal balls and provide perspective on the great non-volatile beyond. They provide insight and analysis on technology trends, disruption, singularities, product roadmaps and completion dates, and other memory issues that may go beyond human predictive capabilities. Bring your opinions, comments or Ouija board, tarot cards, fortune cookies, astrological instruments, tea leaves or magic lamps and join in the discussion. Link to Presentation
Session D-3: Bringing Enterprise-Class Management to Big Memory
The concept of storage at memory speeds is an obvious one. Who wouldn't want storage that is thousands of times faster than an SSD and costs a fraction of DRAM? But how do users manage that storage and carry over essential enterprise-class data services such as data mirroring and snapshots at memory speeds? New virtualization software provides the solution, giving users well-supported access to huge memory lakes along with essential utilities. It will handle the needs of massive databases, real-time analytics, AI model training and execution, financial applications, cybersecurity, AR/VR, cloud computing, and genomics. It also provides the tools needed to make storage at memory speeds practical today.
Introduction to Big Memory
Learn why there are always tiers of memory and storage, the gaps in managing a heterogeneous memory architecture, gaps that are holding back performance, how a memory hypervisor can effectively bridge these gaps and more. Link to Presentation
Big Memory Software
In this presentation, learn about software-defined memory, combining persistent memory and DRAM and get a review of technology MemVerge is developing. Link to Presentation
Implementing Big, Fast Persistent Memory
Discover how we got to persistent memory and how memory storage gaps are being closed today. Link to Presentation
Big Memory for 3D Animation and Visual Effects
Explore big memory’s potential, and some of the work that memory technology has been doing for animated and CG films. Link to Presentation
Session D-4: Top Ten Things You Need to Know about Big Memory Management Today (Panel)
Top Ten Things You Need to Know About Big Memory Management Today
Big memory management will become an important topic for storage designers as persistent memory becomes a more common form of system storage. Implementations have already appeared, and some key issues have become apparent. In particular, the management software must resemble current storage management packages. It must also be scalable, capable of handling enormous amounts of memory, and low overhead, while providing major system utilities managers have come to expect in an enterprise-class form. It must allow for a variety of memory types to coexist in overall systems. Use cases will surely lead to other requirements as they become more widespread. Link to Presentation
Session C-1: How Storage Professionals Should Respond to COVID-19 (Panel)
COVID-19 is everywhere, devastating entire industries such as travel and hospitality, changing everyone's home and workplace, and revolutionizing everyone's lives. How will all these incredible changes affect the storage industry? Who will be the big winners and the big losers? And how must storage professionals change their ways of doing business? TechTarget presents the latest 2020 research on who's buying what, what's driving the purchases, and how the industry can respond best to the ongoing pandemic. The research will also include guidelines for coping with rapid change and massive unpredictability. Link to Presentation
An Introduction to the Storage Networking Industry Association
The Storage Networking Industry Association (SNIA) is a nonprofit organization made up of member companies spanning information technology. This special presentation provides an overview of 2020 work by their 185 member companies in standards, education, and outreach, and offers a glimpse into 2021 plans in compute, memory, and storage activities. Link to Presentation
TechTarget Storage Media: What Sets Us Apart
In this special presentation, TechTarget's VP of Market & Digital Strategy, Jillian Coffin, provides an overview of TechTarget’s global sites devoted to the storage community, the audience -- including audience data around new buying dynamics and more. Link to Presentation