Sorry, you need to enable JavaScript to visit this website.

CMS Summit 2024 Presentation Abstracts

webinar

 Below are the abstracts for the lineup of topics and speakers at our 2024 Compute, Memory, and Storage Summit agenda. 

2024 Compute+Memory+Storage Summit Presentation Abstracts


Opening Remarks

David McIntyre, SNIA CMS Summit Planning Team Co-Chair, SNIA CMSI Marketing Co-Chair; Director, Product Planning and Business Enablement, Samsung Corporation

Abstract

Join David McIntyre as he kicks off our virtual Summit with an overview of the Summit content and presenters, with a focus on the "hot" topics for 2024! 

 

Back to Top


Memories are Driving Big Architectural Changes – Hold Onto Your Hats!

Jim Handy, General Director, Objective Analysis
Tom Coughlin, President, Coughlin Associates

Abstract

Memories, long the slowest-changing of any digital technology, are rapidly springing into new form factors, interfaces, and even core technologies, like MRAM, ReRAM, and FRAM.  This accelerated shift, combined with the computing industry’s sudden adoption of AI, is driving an era of extreme architectural change in systems.  CXL is competing with DDR, HBM is boosting cache sizes, and emerging memories are bringing persistence closer to, and even within the processing chip, while chiplets promise to accelerate that transition.  Meanwhile, new approaches like Processing in Memory (PiM) and endpoint AI are being adopted to increase the amount of information that can be captured and analyzed while reducing the amount of data communicated over the network.  In this presentation, based on a recently-released report, IEEE president Tom Coughlin and noted memory analyst Jim Handy detail the changes the loom before the computing community, and show a likely course for the adoption of new technologies throughout the industry over the next decade and beyond.

Back to Top

 

Storage Requirements for AI

John Cardente, Member of Technical Staff, Dell Technologies

Abstract

While GPUs often steal the limelight, it’s essential to recognize the significant role that storage plays in Artificial Intelligence (AI) infrastructure solutions. Throughout the entire AI lifecycle, from data preparation to pre-training, fine-tuning, checkpointing, and inference, storage systems are critical. They not only keep GPUs busy but also safeguard valuable data. In this presentation, we delve into each of these stages, using concrete examples to highlight common patterns and identify key requirements, particularly related to performance. Additionally, we explore the importance of Deep Learning framework data loader libraries and discuss trade-offs between file-based and object-based storage. By attending this talk, participants will gain a better understanding of AI storage workloads and be better equipped to assess their own infrastructure needs.


Streamlining Scientific Workflows: Computational Storage Strategies for HPC

Dominic Manno, Research Scientist, Los Alamos National Laboratory

Abstract

Scientific simulation in HPC data centers generates and analyzes large datasets to gain insight and test hypotheses. Exploring magnetic reconnection, simulating plasmas flowing over one another, determining particle interactions and trajectories, and simulating an asteroid impact large enough to cause dinosaur extinction are some examples of scientific simulation. These workflows at large scales are time consuming and resource intensive. Single timestep datasets are tens of terabytes and can easily grow into petabyte scale, demanding highly performant storage systems.
 
It often isn’t enough to simply provide high throughput file systems and storage media, instead we must work with domain scientists to understand their workflows then push to gain efficiency throughout the layers of I/O. Computational storage plays a key role in improving this efficiency by providing the capability for function pushdown of storage server operations, data filtering, data indexing, and more. Pairing this alongside popular industry file formats, open-source file systems and protocols, data analytics engines, open scientific datasets, and great industry partners, we have produced multiple proof-of-concepts that show how impactful computational storage can be in scientific simulation. These same principals can be extended to other computing verticals. In this talk we will cover these scientific workflows and how they can benefit from different computational storage configurations.

Increasing AI and HPC Application Performance with CXL Fabrics - A Panel

Kurtis Bowman, Marketing Working Group Co-Chair, CXL Consortium, Moderator
Sandeep Dattaprasad, Sr. Product Manager, Astera Labs, Panelist
Steve Scargall, Sr. Product Manager and Software Architect, MemVerge, Panelist

Abstract

The CXL 3.1 specification introduces CXL fabric manager and extensions, Trusted-Execution-Environment Security Protocol (TSP), and facilitate memory sharing between accelerators and GPUs.
 
This panel presentation will introduce the new features and explore how CXL attached memory to meet the increased memory capacity and bandwidth for HPC, AI, and ML applications in modern data centers. Expert representatives from CXL Consortium member companies will highlight ROI examples when implementing CXL attached memory.
 
Attendees will explore existing and future use cases for CXL and learn more about the products available in the market today.

The Latest on SSD Form Factors 

Jonmichael Hands, Co-Chair, SNIA SSD Special Interest Group

Abstract

Learn how having a flexible and scalable family of form factors allows for optimization for different use cases, different media types on SSDs, scalable performance, and improved data center TCO. We will highlight the latest SNIA specifications that support these form factors, provide an overview of platforms that are EDSFF-enabled, and discuss the future for new product and application introductions.

Enabling an Open Chiplet Ecosystem with UCIe

Brian Rea, Marketing Group Co-Chair, UCI Express and Richelle Ahlvers, Vice Chair and Executive Committee, SNIA

Abstract

UCIe™ (Universal Chiplet Interconnect Express™) is an open specification that defines the interconnect between chiplets within a package, enabling an open chiplet ecosystem and ubiquitous interconnect at the package level. The UCIe specification covers the die-to-die I/O physical layer, die-to-die protocols, and software stack, which leverage the well-established PCI Express® (PCIe®) and Compute Express Link® (CXL®) industry standards.
 
This presentation will introduce the new features in the UCIe 1.1 specification and explore how UCIe addresses customer requests for more customizable package-level integration, connecting best-in-class die-to-die interconnects and protocols from an interoperable, multi-vendor ecosystem.

Ethernet Evolved: Powering AI’s Future with the Ultra Ethernet Consortium

J Michel Metz, Ph.D, Chair, Ultra Ethernet Consortium

Abstract

In the rapidly evolving landscape of Artificial Intelligence (AI) and High-Performance Computing (HPC), the demand for advanced network capabilities has never been greater. The Ultra Ethernet Consortium (UEC) stands at the forefront of this transformation, redefining the boundaries of Ethernet technology to meet these challenges head-on. This presentation will delve into the consortium’s pioneering efforts to adapt Ethernet, traditionally known for its ubiquity and cost-effectiveness, into a high-performance networking backbone capable of supporting the most demanding AI environments.
 
Attendees will learn about:
• Evolving Needs: Characteristics of AI and HPC that necessitate adapting the networking environment, and how to use Ethernet’s ubiquity and flexibility as a powerful solution;
• Scalability and Interoperability: The technical innovations and specifications being developed by UEC to enhance Ethernet’s performance, scalability, and efficiency in AI and HPC applications; and
• Next-Level Performance: Discover how UEC specifications surpass existing Ethernet capabilities, including remote direct memory access (RDMA) and RDMA over Converged Ethernet (RoCE), while supporting strategies for maintaining interoperability and backward compatibility.
 

Computational Storage Case Studies - Real User Deployments 

JB Baker, VP of Product and Marketing, ScaleFlux

Abstract

Computational Storage has been on the hype-cycle for several years now.  With all the smoke, is there really a fire?  Are data centers and enterprises actually deploying computational storage?  If so, are they seeing the reality live up to the hype?
 
JB will touch on some of the promises of computational storage and present several examples of deployments in production environments.  This will include the challenges the users faced, the applications they use, and the benefits they are seeing from deploying computational storage drives.

Back to Top


Breaking Through the Memory Wall with CXL

Ahmed Medhioub, Product Line Manager, Astera Labs

Abstract

Processor performance is rapidly outpacing memory bandwidth, creating a bottleneck or “wall” between compute and memory. This “Memory Wall” limits overall server compute performance, especially in memory-intensive AI applications. This presentation will address how to break through the memory wall with CXL-attached memory. Attendees will learn how popular use cases, such as in-memory databases & AI inferencing, are driving the need for more memory bandwidth and capacity. This session will also introduce how innovations with CXL technology can be used to break through the memory wall to lay the foundation for AI, cloud and enterprise data centers to accelerate computing performance.

Back to Top

GENERATIVE AI: Data architecture with Google AlloyDB & AI

Prasad Venkatachar, Sr. Director Products and Solutions, Pliops

Abstract

This talk will explore the power of Google Cloud’s AlloyDB Omni, a groundbreaking database technology poised to shape 2024’s technological landscape using analytics acceleration and built-in machine learning for optimizing performance and managing complex workloads. This technology enables data platforms for dynamic business analytics and cutting-edge Generative AI (Gen AI) applications, and is compatible with the widely popular PostgreSQL open source database.
 
We will discuss how these platforms are deployed seamlessly across the Edge and Data Center Core on the Lenovo platform, and with Google Cloud for optimizing online transaction processing (OLTP) and real-time data analytics. We'll also delve into the fascinating realm of state-of-the-art Retrieval Augmented Generation (RAG) using Vector Databases using AlloyDB AI feature.

Back to Top

VMware Memory Vision for Real World Applications 

Sudhir Balasubramanian, Sr. Staff Solution Architect, VMware By Broadcom
Arvind Jagannath, Lead Platform Product Manager, VMware By Broadcom

Abstract

VMware has been on an evolving journey on memory innovations mainly first with persistent memory, then with memory tiering, and is now extending that with CXL. CXL provides an opportunity for VMware (by Broadcom) to further improve on performance, and provide further customer benefits such as TCO reduction, server consolidation, and even disaggregation, with increased capacity and bandwidth to run workloads like Mission critical databases, AI/ML  and analytics. Use of accelerators increases the number of use-cases that can be supported with a larger variety of workloads with minimum configuration changes. This session aims to provide real-world application examples using memory tiering.

Back to Top

SNIA Computational Storage Standards

Bill MartinPrincipal Engineer, SSD IO Standards, Samsung Electronics Co., Ltd.
Jason Molgaard, Principal Storage Solutions Architect, Solidigm

Abstract

The SNIA Computational Storage TWG successfully released the Computational Storage Architecture and Programming Model v1.0 in August 2022 and the Computational Storage API v1.0 in October 2023.  The CS TWG continues advancing Computational Storage with enhancements to both the Computational Storage Architecture and Programming Model and Computational Storage API.  The CS Architecture and Programming Model enhancements focus on security for multitenancy and sequencing of commands while the CS API enhancements provide clarifications to facilitate understanding.  This presentation will describe these enhancements in detail and discuss the current state of the SNIA CS Architecture and Programming Model and the current state of the SNIA CS API.

Back to Top

Storage Architecture Optimized for AI Workloads

Paul McLeod, Product Director, Storage, Supermicro

Abstract

Storage optimized for AI workloads must have high performance throughput to stage data on GPU cluster servers while also having a very large, cost-effective, capacity optimized mass storage tier to collect, process and label large data sets needed for the AI model training.  In this session, Supermicro will discuss AI storage solutions using high performance flash-based storage servers and high-capacity disk storage servers with file and object storage solutions from partners such as WEKA, OSNexus, Quantum ActiveScale and Scality.  We will also describe how a 25PB AI-optimized storage implementation was deployed at a leading technology manufacturing company for use in machine vision applications and how similar storage can be deployed for other applications.

Back to Top

 

Edgenuity: The Intersection of Edge and AI

Jeff White, CTO, Edge Product and Operations, Dell Technologies 

Abstract

AI and Edge are two areas that are accelerating and evolving to provide increased utility and transformational capabilities for the enterprise and public sector.  AI has been a prevalent Edge workload since the inception of IoT, and now its role is expanding both in terms of application workloads and how it can enable Edge operations. This discussion will present the intersection of Edge and AI and how it will transform the enterprise. After this session, users will understand:

  • What is a modern edge, how it differs from embedded approaches and why it is important
  • How AI technology is leveraged to build modern edges
  • How AI workloads are supported by the modern Edge

Back to Top

 

Optimizing Content Delivery Network Design

Andy Banta, Storage Janitor, Magnition IO

Abstract

Building Content Delivery Networks, or CDNs, requires distributed, localized collections of compute, memory and storage.  CDNs are built out of groups of servers at a variety of locations with various tiers and types of caches. Modern CDN caches present a huge array of variable configurations.
 
By modularly simulating the components into building blocks, we show how you can quickly try many different sets of configurations.  This modularity also allows plugging in alternative or proprietary components to measure their impact on the overall system cost and performance.  These simulations can be built in days or weeks, instead of the months to years needed to build and test live systems.
 
This session walks through the process of simulating specific CDN configurations. We demonstrate how modular components are swapped and show the simulated performance of different variations. We also demonstrate how to use real-world CDN traces to build realistic scenarios and how the results of the analysis are graphically presented.
 
 
 
 

Creating a Sustainable Semiconductor Industry for the AI Era

Garima Desai, Sustainability Manager, Samsung Semiconductor

Abstract

AI is driving dramatic increases in compute requirements, often combined with equally dramatic power increases. How can we ensure a sustainable approach without losing the innovation potential of AI? This presentation will cover sustainability in the context of semiconductors and AI, drawing on efforts today to reduce carbon emissions and other environmental harms through efficiency improvements, abatement, and recycling. This session will provide a background on sustainability and its relevance today, in addition to the environmental impact of semiconductors. Framing this in the context of the increased demand for semiconductors because of the new AI boom, this presentation will discuss what can be done to help the semiconductor industry move along the green transition.

 

Back to Top

 

A Practical Approach to Device-Level Analytical Offloads

Donpaul Stephens, Founder and CEO, AirMettle, Inc.

Abstract

The odyssey toward enabling high-volume device-level computational storage has been going on since before today’s college graduates were born. But, conflicting requirements of device vendors and application logic have continually kept practical computational storage just over the horizon, as if a mirage.”
 
We will propose a groundbreaking approach that not only simplifies this integration but can accomplish this with only user commands requiring no special privileges while while dramatically reducing the data returned to the host — by over an order of magnitude in many common scenarios. This technique ensures easy implementation on devices and seamless access for higher-level software, leveraging familiar tools like SQL for data analysis without adding complexity to device operations. We will demonstrate how our approach seamlessly aligns with current device APIs, offering a significant leap forward in computational storage.
 
Join us to explore how a reimagined approach to analytical offloads can transform storage devices from mere data repositories into powerful analytical processing engines, marking the end of the computational storage mirage.

NVM Express® Host Managed Live Migration

Mike Allison, Sr. Director NAND Product Planning - Standards, Samsung Semiconductor

Abstract

To minimize any downtime in virtualized and cloud environments, a seamless migration of the Virtual Machine (VM) and associated resources needs to be completed without affecting the user experience in the case of any load balancing, system failures, or system maintenance. When a VM is migrated from one server to another server, the namespaces that the VM has access to also need to be seamlessly migrated. This presentation is an overview of capabilities being investigated by NVMe to support a host controlling the migration of a VM and the namespaces used by that VM to a different controller where that controller may exist in a different server.

Back to Top

Breakthrough in Cyber Security Detection Using Computational Storage

Andy Walls, Chief Architect, CTO, IBM Fellow, IBM Corporation

Abstract

CyberT-attacks, including ransomware are a huge concern for all organizations. There is significant attention in the industry to perimeter security improvements and such things as immutable copies to recover quickly from attacks. Unfortunately, attackers sometimes get in despite our best efforts. Detecting these intrusions quickly is vitally important. It had been thought that there was not much that block storage could do since it may not know the context of the data it is processing. However, IBM with its computational storage device, the FlashCore Module, has figured out how to do AI based ransomware detection within its storage array. This talk will give insight into how it is done and a vision for where we can go from here.

Back to Top

AI: Pushing Infra Boundaries - Memory is a Key Factor

Manoj Wadekar, Hardware Systems Technologist, Meta

Abstract

In recent years, hyperscale data centers have been optimized for scale-out stateless applications and zettabyte storage, with a focus on CPU-centric platforms. However, as the infrastructure shifts towards next-generation AI applications, the center of gravity is moving towards GPU/accelerators. This transition from "millions of small stateless applications" to "large AI applications running across clusters of GPUs" is pushing the limits of accelerators, network, memory, topologies, rack power, and other components. To keep up with this dramatic change, innovation is necessary to ensure that hyperscale data centers can continue to support the growing demands of AI applications. This keynote speech will explore the impact of this evolution on Memory use cases and highlight the key areas where innovation is needed to enable the future of hyperscale data centers.

Back to Top

Bringing Unique Customer Value with CXL Accelerator-Based Memory Solutions

David McIntyre, Director, Product Planning and Business Enablement, Samsung Corporation
Sudhir Balasubramanian, Sr. Staff Solution Architect, VMware by Broadcom
Arvind Jagannath, Lead Platform Product Manager, VMware by Broadcom

Abstract

CXL is mostly talked about in the memory expansion use-case context. However, VMware and Samsung are working together to bring unique value propositions by enabling newer use-cases beyond just memory. A CXL Type-2 accelerator-based solution using a custom hardware-software co-design has the potential to leverage more of the CXL capabilities such as to improve CapEx by reducing TCO, or provide dynamic memory usage and better workload migration performance and thus improving OpEx. We will also describe advantages provided by such an approach and cover real application benchmarks.

Back to Top

Proprietary Interconnects and CXL

Larrie Carr, VP of Engineering, Rambus Inc.

Abstract

As compute architectures expand beyond a single socket, the proprietary interconnect within the architecture becomes part of the solution’s innovation options.   The presentation will look at the history of open interconnects before CXL within the sea of proprietary connectivity and how CXL will most likely co-exist in the future.

Back to Top

 

Smart Data Accelerator Interface: Use Cases, Futures, and Proof Points

Shyam Iyer, Chair, SNIA SDXI TWG; DIstinguished Engineer, Dell Technologies

Abstract

Shyam Iyer, Chair of the SNIA Smart Data Accelerator Interface (SDXI) Technical Work Group, provides an update on this SNIA standard for a memory-to-memory data movement and acceleration interface. Learn about SDXI-based accelerators applicable use cases including those in emerging areas like artificial intelligence, and how you can participate in future work in this growing ecosystem.

Back to Top

 

Security and Privacy Concerns for AI 

Eric Hibbard, CISSP, FIP, CISA, Director, Prodcut Planning-Security, Samsung Semiconductor, Inc.

Abstract

Artificial intelligence (AI) systems are creating numerous opportunities and challenges for many facets of society, including both security and privacy. For security, AI is proving to be a powerful tool for both adversaries and defenders. In addition, AI systems and their associated data must be defended against a wide range of attacks, some of which are unique to AI. The situation with privacy is similar, but the societal concerns are elevated to a point where laws and regulations are already being enacted. This session explores the AI landscape through the lenses of security and privacy.

 

Back to Top

 

Storage Security - Year in Review 

Eric Hibbard, CISSP, FIP, CISA, Director, Product Planning-Security, Samsung Semiconductor, Inc.

Abstract

Attacks against data (e.g., data breaches and ransomware) continue at a dizzying pace, so there is pressure to have storage systems and ecosystems play a more active role in defense. Over the past 9-12 months, there have been promising developments in several industry and standards development organizations that may enhance storage security capabilities. This session summarizes these recent storage security developments, highlights a few important interdependencies, and identifies a few activities that are still underway.

 

Back to Top

 

Storage Sanitization: Recent Evolution

Paul Suhler, Principal Engineer, SSD Standards, KIOXIA

Abstract

The need to eradicate recorded data on storage devices and media is well understood, but the technologies and methodologies to do it correctly can be elusive. A number of new standards build on ISO/IEC 27040 (Storage security) and IEEE 2883-2022 (Standard for Storage Sanitization), and provide clarity on how organizations can evaluate their security needs.

This session will also describe pending revisions of existing standards related to data sanitization, as well as the relationships between the standards developed by various organizations, such as IEEE-SA, ISO/IEC, and NIST.

 

Back to Top

 

Adopting the Zero Trust Paradigm

Eric Hibbard, CISSP, FIP, CISA, Director, Product Planning-Security, Samsung Semiconductor, Inc.

Abstract

The concept of zero trust (ZT)—no trust by default and assume you are operating in a hostile environment—is not new, but applying this concept requires a paradigm shift in the way an organization protects its data and resources. ZT security frameworks typically require users and entities to be authenticated, authorized, and continuously validated before being granted access to applications, systems, and data. Eliminating implicit trust can significantly reduce the exposures from successful attacks.

The U.S. Government has been spearheading adoption of ZT, which is having an impact on the offerings from the security vendor community. Internationally, ZT is gaining traction and is emerging in important security standards. This session highlights important aspects of ZT and provides an update on the state of international activities.

 

Back to Top

 

Overcome Real World Challenges between Data and AI

Steven Yuan CEO, StorageX.ai

Abstract

In an era characterized by exponential growth in data generation, leveraging new infrastructure to manage data-heavy workloads especially Artificial Intelligence (AI) has become essential. 

Today we are going to discuss real world deployment of a new compute & network infrastructure to handle vast amounts of data efficiently and effectively. AI algorithms, particularly those deep learning domain, are designed to process, analyze, and derive insights from large datasets, enabling organizations to make data-driven decisions at unprecedented speed and accuracy.

This topic explores the infrastructure for streamlined data processing, real-time data analytics, and predictive modeling, which significantly reduce the time and effort required to process extensive data volumes with lower latency, demonstrate the transformative impact of AI on data-intensive tasks, highlighting improvements in compute efficiency for decision-making, and predictive accuracy. 

By providing a new perspective of infrastructure to improve process capabilities in handling data-heavy and latency critical workloads, it underscores the potential of AI to revolutionize the data analytics practices and drive innovation across diverse industries.

 

Back to Top