SDC EMEA 2022 Abstracts

 

Plenary Abstracts

 

Enabling Performance and Capacity Scaling for Big Data Using Cloud Storage Acceleration Layer (CSAL)

Mariusz Barczak

Abstract

Cloud service providers are pushing the boundaries of storage capacities to meet the performance requirements of big data workloads and expanding data volumes. Intel and AliCloud collaborated on Cloud Storage Acceleration Layer (CSAL) to address NAND density and scale challenges. CSAL uses Intel Optane SSDs to improve system performance while reducing costs scaling QLC NAND value. In this presentation, we will show and analyze the real world I/O behaviors from big data services of Alibaba Cloud to demonstrate predictable performance using CSAL.


How to quantify the carbon footprint benefits of a computational storage based system

Jerome Gaysse

Abstract

Computational storage is well known for its data movement reduction, leading to higher application performances. The lower power consumption is another benefit which is often highlighted, but a full system life cycle analysis is recommended to get the full picture of the energy used and its related environmental impact.
In this talk, a detailed carbon footprint analysis will be described for a standard system and a computational storage based system. That will help in understanding the computational storage benefits in term of carbon footprint, and will provide some opportunities to pave the way for a roadmap of next generation eco-designed computational storage devices.

LEARNING OBJECTIVES

  • The basis to understand a life cycle analysis
  • Carbon footprint details and benefits for a CS- based system 
  • New design opportunities to reduce the CS carbon footprint 
 

 


Track 1 Abstracts

Container Data Protection: Challenges and Opportunities

Amit Kumar Roushan, Pravin Ranjan & Sushantha Kumar

Abstract

Data protection is one of the key aspects of a stateful application.  In case of containerized deployment scenarios, this is even more crucial considering the ephemeral nature of containers.

This session throws light on key aspects of container data protection, different kinds of challenges involved here. It will also cover the activities going on in this area, solutions being worked out to bridge the gaps

 


Container Storage Observability: State, Challenges and Trends

Sanil Kumar D

Abstract

Data insight is a pivotal area for any technology today. Comprehensive storage monitoring can provide critical health information for preventive and predictive maintenance of various solutions. As the application deployments see an exponential growth towards container environments, container storage monitoring is inevitable. This session will provide the current state of container storage observability, the key challenges in storage monitoring in container deployments (esp Kubernetes), and also the trends. The session will also discuss potential solutions and suggestions based on open source projects for the container observability framework for heterogeneous storage vendors. The session will provide the technical architecture for a Kubernetes native observability solution and discuss the possibilities.

LEARNING OBJECTIVES

  • Introduce Storage Monitoring in Kubernetes
  • The current challenges and solutions for container storage monitoring 
  • How to build kubernetes native observability 
  • How to handle heterogeneous storage monitoring

Is Your Storage Software ready for 80M IOPS?

Karol Latecki and Konrad Sztyber

Abstract

Since its inception, the Storage Performance Development Kit (SPDK) project has demonstrated that it's possible to build efficient storage systems capable of tens of millions of IOPS. This talk will describe the techniques and principles used to achieve over 80M IOPS at an amazing average latency of just 25 microseconds. We will also discuss the workloads, test methodologies, and automation tools that allow for continuously measuring and improving performance of NVMe, NVMe-oF and Vhost implementations.

LEARNING OBJECTIVES

  • Understand which design concepts SPDK follows to achieve high storage performance and how to leverage this in you storage stack
  • Understand how continuous benchmark tests help SPDK verify it's performance
  • Understand how SPDK approach impacts CPU efficiency and performance as compared to Kernel counterparts

Ansible Automation Mesh to Connect Disparate Automation Components for Hybrid Clouds

Kannadasan Palani

Abstract

Ansible has made automation frameworks to seamlessly function with multi-cloud and hybrid cloud environments. With auto-scaling and configuration management, Ansible-based automation platforms help CI/CD developers with network compliance, SaaS operations, and end-to-end legacy systems modernization. The challenge is to manage the Docker and Containers for Hybrid and Multi-Cloud environments, which are increasingly gaining preference because of their flexibility, performance, and economy. How can one ensure an automation framework for application deployment and configuration management for DevOps teams working with remote cloud instances and niche digital architectures? This paper aims to answer these questions by demonstrating Ansible’s contribution to the different phases of the CI/CD pipeline. The paper will also go to lengths to describe Ansible Automation Mesh, where its playbooks deal with configuration management automation to ensure rapid development and zero downtime for Hybrid and Multi-cloud environments.

LEARNING OBJECTIVES

  • Need for robust Automation Frameworks to overcome the challenges posed by Hybrid and Multi-Cloud environments 
  • Role of Ansible to implement scalable automation for end-to-end configuration management, along with enabling VM to Docker migration
  • Working with Ansible Playbooks to ensure auto-scaling application deployment different phases of the CI/CD pipeline

An Innovative Approach to Achieve Compositionality using Multi-Version Object Based Transactional Systems to Exploit multi-cores

Archit Somani and Sweta Kumari

Abstract

In the modern era of multi-core systems, utilizing cores is a tedious job. Synchronization and communication among processors involve high cost. Software transaction memory systems (STMs) addresses these issues and provide better concurrency in which programmer need not have to worry about consistency issues. Another advantage of STMs is that they facilitate compositionality of concurrent programs with great ease. Different concurrent operations that need to be composed to form a single atomic unit is achieved by encapsulating them in a single transaction. 

We introduce an innovative STM system to exploit multi-cores as multi-version object based STM (MVOSTM) which works on higher level operations such as insert, delete instead of read, write operations; ensures compositionally by making the transactions atomic; harness the greater concurrency while maintaining multiple versions; and reduces the number of aborts further. We have used MVOSTM for the list and hash-table data structure as list-MVOSTM and HT-MVOSTM. Experimental evaluation shows that list-MVOSTM and HT-MVOSTM outperform state-of-the-art STMs and incur negligible aborts.

LEARNING OBJECTIVES

  • Insights to exploit multi-core systems using an inno- vative framework MVOSTM
  • How to ensure the compositionality efficiently?
  • How to achieve greater concurrency and minimize the aborts?
  • Applicable for Distributed Transactional Memory sys- tems, Storage, Blockchain, Persistent Memory, etc.

Unlocking the potential of NVMe-oF in Modern Storage Arrays  

Constantine Gavrilov

Abstract

During the last 10 years, a new communication protocol to interface with SSDs has been specified from the ground-up, allowing to fully exploit the levels of parallelism and performances of all-flash storage: NVMe, and NVMe-oF. NVMe-oF promises to enable the performances of direct attached all-flash storage with the flexibility and TCO savings of shared storage. Implementing NVMe-oF support in enterprize storage arrays allows to combine NVMe protocol performance with the reach feature set of modern storage arrays and enables modern storage arrays to meet growing customer demands.

LEARNING OBJECTIVES

  • Architecturing NVMeOF Solutions
  • NVMeOF RDMA Solution
  • NVMeOF TCP Solution
  • NVMeOF Performance

Machine Learning at Scale on Container (Kubernetes) and distributed file systems  

Suman Debnath

Abstract

Data science, machine learning (ML), and artificial intelligence have exploded in popularity in the last few years, with companies building out dedicated ML teams. Kubeflow is an OpenSource, ML toolkit for Kubernetes which provides useful components resolving problems in multiple areas. In this hands on session we will discuss about how you can make a scalable architecture for ML training and inference at scale using a distributed storage in a backend and Kubeflow.

LEARNING OBJECTIVES

  • Distributed ML training using Kubeflow 
  • How Kubeflow works with a persistent and scalable file system
  • Deploy an end to end ML workflow

Persistent Memory usage model in Ceph Crimson  

Chunmei Liu and Yingxin Cheng

Abstract

Crimson is a project to refactor ceph osd based on the shared-nothing seastar framework. The current Crimson SeaStore is mainly targeting NAND SSD because of their high performance-price ratio and is heavily optimized for random read and sequence write patterns. This presentation will extend the SeaStore framework to support multiple tiers of devices, mainly byte addressable random access devices such as optane Persistent Memory (PMEM), and random block access devices such as optane SSD.
PMEM has the following usage models: 1) As transient cache which has nearly the same speed and latency of DRAM but has a much larger size; 2) As persistent storage for metadata which is updated synchronously like memory; 3) As a hot device tier for both metadata and data. In the extended SeaStore framework, each kind of device has its own device manager to arrange the usage of the device: 1) NAND SSD uses segment manager which partition device into segments and uses a cleaner to reclaim space; 2) Optane SSD uses random block device manger which uses bitmap to allocate space; 3) and PMEM uses its own manager which also uses bitmap to allocate space. Some common paths with different devices are shared, such as transaction commit and delta processing, with adjustments needed to adapt to the specialized I/O patterns.

LEARNING OBJECTIVES

  • Get to understand the architecture of Crimson SeaStore 
  • Get to understand multi-tier device support in SeaStore 
  • Get to understand the data work flow in SeaStore

Persistent Memory and NAND Flash: Where to use and why?  

Pekon Gupta

Abstract

This paper discusses similarities and differences between Persistent Memory and NAND Flash storage technologies.
Media specific characteristics like byte address ability, access modes, and cost determine where Persistent Memory is better fit than NAND Flash Storage. With adoption of serial and fabric attached protocols like CXL, Gen-Z and OpenCAPI, Persistent Memory can be packaged in same form-factor as NVMe drives, and it can reuse existing NVMe management infrastructure. This helps OEM to develop common chassis architecture which can house both Memory and NAND Flash SSDs in same slots.

This paper highlights few of the ideas where Persistent Memory and NAND Flash drives become physically replaceable.

LEARNING OBJECTIVES

  • Difference between Memory, Persistent Memory and NAND Flash
  • Different Persistent Memory solutions (NVDIMM-N, NVDIMM-P, NVDIMM-F)
  • Common form-factors between 'serial attached Persistent Memory' and NVMe SSD
  • Common management interfaces between 'serial attached Persistent Memory' and NVMe SSD

Data Protection and Rapid Recovery for Peta Scale Cloud Deployments without any compromise  

Ofer Frishman

Abstract

Some things can’t scale in the cloud. When you are trying to get all the performance out of your systems for SaaS and IaaS instances and RAID 0 seems like the option. What do you do when you have 60 servers go down due to an SSD failure? In this session, you will learn about new breakthrough data protection technology for SSDs that gives you better performance during a drive rebuild than you can get from RAID 0 while getting RAID 5 like data protection.

LEARNING OBJECTIVES

  • Learn about efficient ways to protect NVMe drives while not comprising performance
  • Learn about how CPU deficiencies can be overcome with NVMe flash
  • Learn about modern ways to handle data destined for NVMe flash

Kerberos/Authentication Updates in Samba  

Stefan Metzmacher

Abstract

On the domain controller side we got a lot of updates recently:
- Updated Heimdal
- Working with the latest MIT Kerberos
 

On the member server side we fixed some critical bugs and have plans for future improvements how a file server can avoid as much domain controller interaction as possible.

LEARNING OBJECTIVES

  • How Samba plans to use Kerberos FAST
  • How you can reliable change a machine password
  • Why it is so important to behave as exactly identical as possible compared to a Windows server

Benefits of Compute Express Link™ (CXL™) for Next-Generation Datacenters  

Danny Moore

Abstract

Compute Express Link™ (CXL™) is an open industry-standard interconnect offering coherency and memory semantics using high-bandwidth, low-latency connectivity between the host processor and devices such as accelerators, memory buffers, and smart I/O devices. CXL technology is designed to address the growing needs of high-performance computational workloads by supporting heterogeneous processing and memory systems for applications in Artificial Intelligence, Machine Learning, communication systems, and High-Performance Computing. These applications deploy a diverse mix of scalar, vector, matrix, and spatial architectures through CPU, GPU, FPGA, smart NICs, and other accelerators. The CXL 2.0 specification introduces support for switching, memory pooling, and persistent memory – all while preserving industry investments by supporting full backward compatibility. Based on member feedback, the CXL Consortium technical working groups published Engineering Change Notices (ECN) for the CXL 2.0 specification, primarily to enhance performance, reliability, software interface, and testability while offering design simplification. The presentation will also provide a high-level sneak peek of the latest advancements in the CXL 3.0 specification development, its new use cases and industry differentiators.

LEARNING OBJECTIVES

  • Learn how CXL supports dynamic multiplexing between a rich set of protocols that includes I/O, caching and memory semantics 
  • Explore how CXL addresses address the growing needs of high-performance computational workloads 
  • Introduce new features in the CXL 2.0 ECNs 
  • Gain insight into what’s coming up next in the CXL 3.0 specification development 
 

A Persistent CXL Memory Module with DRAM Performance 

Bill Gervasi

Abstract

Data persistence on CXL is an essential enabler toward the goal of instant-on processing. DRAM class performance combined with non-volatility on CXL enables a new class of computing architectures that can exploit these features and solve real-world bottlenecks for system performance, data reliability, and recovery from power failures. New authentication methods also enhance the security of server data in a world of cyberattacks.

LEARNING OBJECTIVES

  • CXL challenges and advantages 
  • Persistent main memory vision
  • Security advantages of a robust system management channel
 

Accelerating Oracle Workloads using Intel DCPMM, Oracle PMEM Filestore on VMware 

Sudhir Balasubramanian and Arvind Jagannath

Abstract

Enabling, sustaining and ensuring the highest possible performance along with continued application availability is a major goal for all mission critical applications to meet the demanding business SLA’s., all the way from on-premises to VMware Hybrid Clouds. 
Please join this session to learn how performance of business-critical Oracle workloads can be accelerated via Persistent Memory technology using Intel Optane DC PMM in App Direct mode backed Oracle 21c Persistent Memory Filestore on VMware vSphere platform, thereby achieving those stringent business goals and enhancing performance.

LEARNING OBJECTIVES

  • High level overview of Persistent memory and how VMware vSphere is able to consume it
  • High level overview of Oracle 21c Persistent Memory Filestore using Intel Optane DC PMM in App Direct mode on VMware vSphere 
  • How performance of business-critical Oracle workloads can be accelerated via Persistent Memory technology

WineFS: a hugepage-aware file system for persistent memory that ages gracefully 

Rohan Kadekodi

Abstract

Modern persistent-memory (PM) file systems perform well in benchmark settings, when the file system is freshly created and empty. But after being aged by usage, as will be the normal mode in practice, their memory-mapped performance degrades significantly. This paper shows that the cause is their inability to use 2MB hugepages to map files when aged, having to use 4KB pages instead and suffering many extra page faults and TLB misses as a result.

We introduce WineFS, a novel hugepage-aware PM file system that largely eliminates this effect. WineFS combines a new alignment-aware allocator with fragmentation-avoiding approaches to consistency and concurrency to preserve the ability to use hugepages. Experiments show that WineFS resists the effects of aging and outperforms state-of-the-art PM file systems in both aged and un-aged settings. For example, in an aged setup, the LMDB memory-mapped database obtains 2× higher write throughput on WineFS compared to NOVA, and 70% higher throughput compared to ext4-DAX. When reading a memory-mapped persistent radix tree, WineFS results in 56% lower median latency than NOVA.

LEARNING OBJECTIVES

  • Trade-offs of different file system design choices for persistent memory
  • Effects of file systems on memory mapped applications
  • Importance of huge pages in the performance of memory mapped applications
  • Effect of aging on persistent memory file systems

Fabric Notifications - An Update from Awareness to Action 

Howard Johnson

Abstract

Introduced in 2021, Fabric Notifications developed into a key resiliency solution for storage networks. Providing the ability to respond to events impacting the Fabric, Fabric Notifications enhances the overall user experience when deploying traditional Fibre Channel SANs. In this session, the improvements in the user experience are profiled for solutions using Linux, AIX, and PowerPath. The latest developments in both standards and market solutions are also provided as an update to this exciting new technology.

LEARNING OBJECTIVES

  • Provide an update of Fabric Notifications as defined by the Fibre Channel standards.
  • Describe the mechanisms and techniques for implementing Fabric Notifications.
  • Profile specific deployments of Fabric Notifications solutions.

 


Track 2 Abstracts

SSD Thermal Impact Study with Different Workloads

Ganesh Babu Vasudevan, Mohan Raj Veerasamy and Javier Gonzalez

Abstract

The Solid State Drive (SSD) capacity and performance is increasing over the years and due to this thermal management in SSD is becoming very challenging. As SSD works harder to meet growing demands, it consumes more power and consequently generates more heat. Apart from designing the SSDs to tolerate higher temperature, another effective solution to address the overheating issue is to have a good thermal management. The thermal throttling mechanism is thus fundamental for SSDs. A thermal sensor is used in the drive to monitor the temperature and temperature details can be extracted via S.M.A.R.T. command.

The idea of this paper is to Study thermal impact of different types of IOs, chunk size in multi core SSD with different density, Die Size and NAND Type (SLC & TLC) as parameter. This study will help us in understanding & improve the thermal throttling policy.

LEARNING OBJECTIVES

  • The temperature saturates after certain point of time for all types of workload. 
  • The temperature increase is less for higher chunk size IOs compare to lower chunk size. 
  • The Operating temperature in current SSDs are very less compared to previous Era SSD’s which exhibits higher temperature.

Bringing certainty to Enterprise disk-drive failure management using Conformal Prediction

Hemant Gaikwad & Rahul Vishwakarma

Abstract

The traditional methods of managing enterprise disk drives leverage multiple machine learning models which include its inherent bias or uncertainties. However, recent development and growing interest of academia and industry in Conformal Prediction has further led to newer perspectives on solving the existing problems with the machine learning models. Conformal prediction (CP) is a set of algorithms devised to assess the uncertainty of predictions produced by a machine learning model. 
In this talk, we will first introduce Conformal Prediction, followed by a comparative analysis of solving the problems (using the traditional machine learning method and conformal prediction framework). We will discuss three issues of disk drive management using conformal classification – 1) disk drive failure estimation, 2) inventory planning of individual storage drives, and 3) mechanism for ranking based dynamic hot-sparing. 
To conclude, we will discuss in the application section, how the framework can be used for ransomware detection and designing an efficient disk drive scrubbing mechanism.

LEARNING OBJECTIVES

  • Introduction to Conformal Prediction 
  • Comparative analysis of solving a problem using traditional machine leaning method and conformal framework 
  • Application of the framework – Novel method of ranking based dynamic hot-sparing and efficient disk drive scrubbing mechanism

Symlink races for dummies and how to deal with them

Volker Lendecke

Abstract

This talk will in detail describe the symbolic link race condition that led to the Samba CVE-2021-20316. Symlink races lead to CVEs all over the place, even the secure programming language Rust was recently bitten by it with CVE-2022-21658. Samba solved the problem with a major development effort over many months, however the solution comes with a performance penalty. In the best tradition of "Make it right first, make it fast", work is ongoing to refactor Samba's internals to not only get back the speed of the insecure path handling, but potentially go beyond that and moreover add proper SMB2 symlink error handling. The talk will describe Samba's current solution, the reasons why we did not solve it "the right way" from the beginning, and our way out of this.

LEARNING OBJECTIVES

  • Security problems due to symbolic links 
  • Making path-based operations safe in Posix 
  • Performance characteristics of secure Posix calls

Samba, quo vadis? 

Ralph Bohme

Abstract

This talk is going to give an overview of recent changes in the Samba fileserver and an outlook on the development roadmap. Recent development has been mainly focusing on security resulting in the release of Samba 4.15 last year.

Looking forward there are many things the Samba fileserver development team has on its todo list and this presentation will give a first hand insight into the making of the next Samba versions.

LEARNING OBJECTIVES

  • Samba fileserver updates
  • Samba roadmap

 


The importance of QoS latency on application performance

Or Lapid

Abstract

As new SSD architectures are achieving 1M IOPS for Random Operations, a “one in a million” event happens every second. This puts the term Quality of Service (QoS) as significant data point for SSD performance benchmarks and how they impact application performance.

LEARNING OBJECTIVES

  • Explanation of QoS term and why it matters to applications.
  • Benchmark QoS data for 5 9’s and 6 9’s.
  • Demonstrate the effect of QoS on a RocksDB Use Case.

You thought SPDK was everywhere before... now in infrastructure offload too!

Tomasz Zawadzki and Jacek Kalwas

Abstract

Infrastructure offload based around NVMe-oF can deliver performance of direct attached storage with the benefits and composability of shared storage. Storage Performance Development Kit (SPDK) is a set of drivers and libraries uniquely positioned to demonstrate how projects like Infrastructure Programmer Development Kit (IPDK) can provide vendor agnostic high performance and scalable framework. This session will discuss how the SPDK NVMe-oF target has evolved to enable infrastructure offload.

LEARNING OBJECTIVES

  • Understand what Infrastructure Processing Unit (IPU) is and interaction between its storage components
  • Understand how NVMe-oF transports can implement device specific interaction to emulate NVMe devices
  • Understand how infrastructure offload impacts orchestration using Live migration as an example

Accessing files from the smallest devices to the largest, and to the cloud: Improvements to SMB3.1.1 and Linux

Steve French

Abstract

Accessing files securely and efficiently matters, and over the past year many improvements have been made in Linux for accessing files remotely via SMB3.1.1. This has been a great year for cifs.ko with the addition of new SMB3.1.1 features and optimizations. It continues to be the most active network/cluster file system on Linux. And now with the addition of a kernel server to Linux (ksmbd), there are multiple Linux server options (Samba and ksmbd). 

Improvements to performance have been made by adding support for handle leases (deferred close), better optimizing multichannel, and by changes to read ahead caching, and directory and metadata caching and also signing improvements have been made. Offline caching (fscache) has been rewritten and improved, and support for the Witness protocol (server notification about key events like server moving), and security has improved with support for the strongest encryption, and more recently the exciting work on QUIC. This presentation will go through the features added over the past year to the Linux client (and kernel server) and demonstrate how they help common scenarios, from accessing the cloud (like Azure) to accessing Samba, Windows, Macs and the new Linux kernel server (ksmbd).

LEARNING OBJECTIVES

  • What new SMB3 features for accessing files remotely from Linux have been added in the last year? 
  • What SMB3.1.1 improvements are expected in the coming year to allow for more efficient access to remote files? 
  • What configuration and mount options should be considered for various common scenarios? 
  • What is the status of the new kernel server for Linux (ksmbd) and what can it be used for?

What’s New in SNIA Swordfish™

Richelle Ahlvers

Abstract

The SNIA Swordfish™ specification for the management of storage systems and data services is an extension of the DMTF Redfish® specification. Together, these specifications provide a unified approach for the management of servers and storage in converged, hyper-converged, hyperscale, and cloud infrastructure environments.
If you’ve heard about this latest wave in storage management, but are looking for the latest updates to the functionality of Swordfish, this presentation is for you. The speaker will catch up on the standard’s most recent developments, which include:

  • Expanded support for NVMe and NVMe-oF Devices
  • Managing Storage Fabrics
  • Standardized Capacity and Performance Metrics Management

LEARNING OBJECTIVES

  • Provide the audience with details on how the Swordfish specification and ecosystem has been updated
  • Describe how the latest Swordfish updates can improve storage management.
  • Give a sneak peek of the next steps in the Swordfish development roadmap

Kubernetes and Fibre Channel: A Compelling Case

Mark Jones

Abstract

Over the past decade IT organizations have discovered the benefits of moving applications and storage to the cloud. Recent trends point to an increase of on premises hybrid cloud solutions. Cloud-native applications running in containers has driven the need for high-performance persistent storage that is both easy to deploy and manage. 

In this 20-minute presentation, Fibre Channel expert Mark Jones will explore the role that Fibre Channel plays in merging the benefits of a cloud-based infrastructure with the advantages of an on-premises hosting solution for gains in reliability, performance, and scalability – all with cost reduction. With a focus on Kubernetes, Mark will explain how these best practice environments can be combined for the next generation of application deployments.

LEARNING OBJECTIVES

  • How organizations are deploying cloud
  • Introduction to containers and persistent storage
  • The role of Fibre Channel in Kubernetes

Smart Contract for DNA based archival storage

Rahul Vishwakarma & Hemant Gaikwad

Abstract

Digital information storage in DNA is not a new idea, and the recent developments in the synthesis of DNA have significantly reduced the price per base pair. The current price per base-pair remains $0.07, far from commercial implementation compared to HDD or SSD based storage cost. Earlier work focuses on designing a data encoding method for DNA-Based storage, error correction code, genomic compression, enzymatic method of synthesizing nucleotides in the lab, and few works on encrypting the data. Unfortunately, little work has been done to reduce the amount of nucleotide required for digital information representation, detect malicious tampering of digital data, and choose the deduplication method as compression doesn’t help much for genomic data.

This talk will address the challenges encountered and present proposed solutions for reducing the storage cost for DNA-Based storage per unit of digital information, which uses similarity-based variable-length deduplication with delta encoding).

We also discuss storing archival storage metadata in Blockchain, making data immutable and tamper-proof (smart contract for similarity-based deduplication) and eliminating a single point of failure (decentralized meta-data storage in Blockchain). To conclude, we will present an evaluation of considering the Polysaccharide (Sugar) as one of the archival storage media and comparative analysis with DNA-based storage.

LEARNING OBJECTIVES

  • Storage of archival digital data on DNA 
  • Designing a smart contract based on deduplication 
  • Leveraging Blockchain as metadata for archival storage

How is 64GFC different?

Fausto Vaninetti

Abstract

Demand for higher bandwidth to access data in a secure and reliable way keeps growing. This has resulted in a transition toward the most recent implementation of Fibre Channel technology operating at 64G speed. Despite being fully backward compatible, the new solution introduces key changes from previous generations. In this session we discuss the new modulation schema, the enhanced FEC algorithm, the modified link bring up sequence and other technical considerations.

LEARNING OBJECTIVES

  • What is driving need of 64G FC 
  • How is 64G different than previous speeds 
  • How is the new modulation schema affecting the implementation

Computational Storage Drive: Making the Best Data Structure B+ Tree Even Better

Tong Zhang

Abstract

This proposed talk will present a design solution that improves the speed performance and cost efficiency of B+ tree by leveraging computational storage drives (CSDs) with built-in transparent compression. As the most widely used indexing data structure, B+ tree powers almost all the relational database management systems today. Nevertheless, it is well known that B+ tree has two major shortcomings: low storage efficiency and high write amplification. The arrival of CSDs with built-in transparent compression straightforwardly mitigates the storage efficiency issue for B+ tree. This proposed talk will show that CSDs can also be leveraged to significantly reduce the B+ tree write amplification and hence improve its speed performance. The key is to exploit the fact that CSDs with built-in transparent compression allows data management software employ sparse data structure without sacrificing the true physical storage cost. Experimental results show that the developed solution can reduce the B+ tree write amplification by over 10x.

LEARNING OBJECTIVES

  • Resolve misconception and over-hypes around computational storage
  • Understand the rationale behind the innovating data management software with computational storage drive
  • Learn the specific design techniques that enable relational database take full advantage of computational storage drive
 

PCIe® 6.0 Specification: Advancing Artificial Intelligence and Machine Learning Applications

Vamshi Kandalla and Miki Takahashi

Abstract

For the past three decades, PCI-SIG® has delivered a succession of industry-leading PCI Express® (PCIe®) specifications that remain ahead of the increasing demand for a high-bandwidth, low-latency interconnect for compute-intensive systems in diverse market segments, including data centers, artificial intelligence and machine learning (AI/ML), high-performance computing (HPC) and storage applications. 

In early 2022, PCI-SIG released the PCIe 6.0 specification to members, doubling the data rate of the PCIe 5.0 specification to 64 GT/s (up to 256 GB/s for a x16 configuration). To achieve high data transfer rates with low latency, PCIe 6.0 technology adds innovative new features like Pulse Amplitude Modulation with 4 levels (PAM4) signaling, low-latency Forward Error Correction (FEC) and Flit-based encoding. PCIe 6.0 technology is an optimal solution to meet the demands of Artificial Intelligence and Machine Learning applications, which often require high data bandwidth, low latency transport channels.

This presentation will explore the benefits of PCIe 6.0 architecture for AI/ML workloads and its impact on next-generation cloud data centers. Attendees will also learn about the potential AI/ML use cases for PCIe 6.0 technology.

LEARNING OBJECTIVES

  • Benefits of PCIe 6.0 architecture for Artificial Intelligence & Machine Learning (AI/ML) workloads
  • PCIe 6.0 technology impact on next-generation cloud data centers
  • Potential Artificial Intelligence & Machine Learning (AI/ML) use cases for PCIe 6.0 technology

Why SAS is a Gold Mine for Crypto

Jonmichael Hands & Cameron Brett

Abstract

Just as with mining in the real world, the people who mine cryptocurrency use powerful equipment to increase their chance of finding valuable resources. But all that power comes at a cost: namely, massive energy consumption. However, next generation blockchains such as the Chia Network could change all that. Unlike other cryptocurrencies, you don't mine Chia with a processor, graphics card, or an ASIC miner. Instead, you farm Chia with storage space, which is where hard drives and solid-state drives come in, using the Serial Attached SCSI (SAS) infrastructure to connect them. In this presentation, Jonmichael Hands, vice president, Storage Business Development, Chia Network, will explain the “proof of space and time” (storage capacity) concept that is the foundation of the Chia network, and why SAS is a great technology for farming Chia. He will also discuss the difference between Chia plotting and farming and how they are carried out with the use of storage devices.

The presentation will be moderated by Cameron Brett, president of the SCSI Trade Association (STA) and director of Enterprise and Cloud Storage Marketing at KIOXIA America. SAS is the leading storage interface used in enterprise servers and storage deployments in data centers.

LEARNING OBJECTIVES

  • What the “proof of space and time” concept is
  • The difference between Chia plotting and farming 
  • Which SAS products can help farm for Chia

Computational Storage: How do NVMe CS and SNIA CS work together? 

Bill Martin

Abstract

NVMe and SNIA are both working on standards related to Computational Storage. The question that is continually asked is are these efforts are compatible or at odds with each other. The truth is that many of the same people are working on both of these standards efforts and are very interested in ensuring that they work together as opposed to conflicting with each other. This presentation will discuss how the two standards efforts go hand in hand, the aspects of the SNIA API that support the NVMe efforts, and the NVMe efforts to share common architectural structures with those defined in SNIA. As part of this discussion, a lexicon of terminology used in both organizations will be presented and the decoder ring that allows you to understand one document in terms of the other in spite of some differences in names used.

LEARNING OBJECTIVES

  • Discuss the similarities between SNIA and NVMe work on Computational Storage
  • What are the components of SNIA CS Architecture 
  • What are the components of NVMe CS
 

Automotive SSD Specification Update 

Bill Gervasi

Abstract

The next generation of automobiles moves to the adoption of PCIe for data communications in vehicles, and the JEDEC Automotive SSD solution enables a high performance, high reliability solution for this shared centralized storage. Features such as SR/IOV highlight the requirements of these computers on wheels with multiple SoC functions for vehicle control, sensors, communications, entertainment, and artificial intelligence.

LEARNING OBJECTIVES

  • Solid state storage for automotive applications 
  • Options for product differentiation 
  • Common footprint enabling multi-sourcing

 

 

How to quantify the carbon footprint benefits of a computational storage based system

Jerome Gaysse

Abstract

Computational storage is well known for its data movement reduction, leading to higher application performances. The lower power consumption is another benefit which is often highlighted, but a full system life cycle analysis is recommended to get the full picture of the energy used and its related environmental impact.
In this talk, a detailed carbon footprint analysis will be described for a standard system and a computational storage based system. That will help in understanding the computational storage benefits in term of carbon footprint, and will provide some opportunities to pave the way for a roadmap of next generation eco-designed computational storage devices.