2020 SDC EMEA Abstracts

Plenary Abstracts

 

Enabling the journey to the Hybrid Multi Cloud with Storage Technology

Andy Walls

Abstract

Clients tell us their top priorities are consistent: they are looking to store more data and apply AI and big data solutions to gain faster business insights; they are deploying applications in Red Hat OpenShift and other container environments to improve hybrid multicloud application and data mobility, and they need to be able to do all of this quickly and easily in a heterogeneous environment with constrained budgets. IBM Storage provides solutions that simply address these modern data needs from creation to archive with industry first technology: new standards for performance, new standards on capacity, simplified hybrid cloud connectivity. Learn how to make storage simple, faster and better for hybrid multicloud.


Zoned Namespaces: Standardization and Linux Ecosystem

Javier Gonzalez

Abstract

Architectures implementing Open-Channel SSD concepts have successfully made their way into the data center. However, the lack of a standard has facilitated fragmentation and, in turn, hindered sound adoption. To address this problem, NVMe is standardizing Zoned Namespaces - a new namespace type that allows applications to (i) make explicit decisions around data placement on areas of the device (zones) and (ii) have control over zone states, including reset cycles. The result is lower write- and space-amplification, lower cost and higher predictability. In this talk, we will discuss the concepts around ZNS, the efforts being made to add support for it in Linux, and the user-space ecosystem to enable suitable applications (e.g., RocksDB). Also, we will cover some of the features we are adding to ZNS in order to further reduce tail latencies and provide QoS.

 

 
 

Track A Abstracts

Completing the Memory and Storage Hierarchy

Eran Zwilling

Abstract

The rapid increase in data creates gaps within the memory and storage hierarchy, as it struggles to keep up the pace. With the march of technology, we describe these gaps and address them. This presentation will review the strategy and ways to close these gaps by leveraging 3D XPoint™ and 3D NAND technologies.

 


 

PMEM File-system in user-space

Shachar Sharon

Abstract

Developing a modern file-system over PMEM device is an uncharted territory challenge. We present some of the lessons learned at NetApp from developing an enterprise grade PMEM based file-system. We shall also introduce an open-source re-write of PMFS in user-space over ZUFS, including deep-dive into why and how to do persistent data-structures layout.

Learning outcomes

1. Insights on developing PMEM based file-system

2. Do and don't when working with PMEM

3. Persistent data-structures how-to


 

Linux block integrity in building and testing storage systems

Mikhail Malygin

Abstract

Linux block integrity is a well-known block layer subsystem that helps to detect and prevent data corruption. This talk is based on hands on experience in building and testing storage systems and provides solutions for challenges faced in the block integrity stack. It also covers specifics of integrity implementation in SCSI or NVMe kernel drivers as well as in virtual environments: qemu, virtio, vhost.

Learning outcomes

1. Linux Block integrity

2. SCSI/NVME PI

3. Block integrity virtualisation


 

Understanding NVMe namespaces

Or Lapid & Esther Spanjer

Abstract

In this session we will describe the details of NVMe namespaces, what they are and what they do. We will show how they are technologically very different from conventional LUNs and partitions (controller or operating system based). We’ll describe two acceleration modes namespaces enable (dedicated and shared) as well as show an example of their use in tiered storage. 
 

Learning outcomes

1. Dedicated, server-internal storage acceleration

2. Shared, remote platform acceleration via high-bandwidth fabrics

3. Namespaces and orderly IO benefits

4. Using namespaces to accelerate applications with fixed (limited) thread counts


Shared storage using NVMe-oF

Murali Rajagopal and Madhu Pai

Abstract

This session provides a description of VMware’s support for shared storage using NVMe-OF. VMware’s storage partner NetApp will provide an update on their implementation of NVMe-OF using Fibre Channel as the transport protocol

Learning outcomes

1. Provides an update on current OS support for NVMe-oF

2. Provides an update on current target support for NVMe-oF from NetApp

3. Provides a description of how Fibre Channel is used as the transport for NVMe-oF


Apps can quickly destroy your mobile's flash: Why they don't, and how to keep it that way

Aviad Zuck

Abstract

Although flash cells wear out, a typical SSD has enough cells and sufficiently sophisticated firmware that its lifetime generally exceeds the expected lifetime of its host system. Even under heavy use, SSDs last for years and can be replaced upon failure. On a smartphone, in contrast, the hardware is more limited and we show that, under heavy use, one can easily, and more quickly, wear out smartphone flash storage. Consequently, a simple, unprivileged, malicious application can render a smartphone unbootable (“bricked”) in a few weeks with no warning signs to the user. This bleak result becomes more worrisome when considering the fact that smartphone users generally believe it is safe to try out new applications. To combat this problem, we study the I/O behavior of a wide range of Android applications. We find that high-volume write bursts exist, yet none of the applications we checked sustains an average write rate that is high enough to damage the device (under reasonable usage assumptions backed by the literature). We therefore propose a rate-limiting algorithm for write activity that (1) prevents such attacks, (2) accommodates “normal” bursts, and (3) ensures that the smartphone drive lifetime is longer than a preconfigured lower bound (i.e., its warranty). In terms of user experience, our design only requires that, in the worst case of an app that issues continuous, unsustainable, and unusual writes, the user decides whether to shorten the phone’s life or rate limit the problematic app.
 

Learning outcomes

1. Mobile storage devices are vulnerable to wear out attacks

2. Benign apps can prematurely wear out mobile devices

3. Understanding app I/O behavior is critical to premature wear out defenses


Architecture Development with Computational Storage

Eli Tiomkin

Abstract

With the onset of the Computational Storage TWG and growth of interest in the market for these new and emerging solutions, it is imperative to understand how to develop, deploy and scale these new technologies. This session will walk through the new definitions, how each can be deployed and show use cases of NGD Systems Computational Storage Devices (CSD).

Learning outcomes

1. Learn about what a CSD is and why it matters

2. See new ways to eliminate data movement while gaining value of data

3. See how easy the deployment of Computational Storage can be


NVMe device emulation using a SmartNIC

Idan Burstein

Abstract

An IPU (IO processing unit) is a smart NIC with on-board processors that can run software. An IPU device presents many new opportunities combining the flexibility of software in the network path. An exciting opportunity for IPUs is NVMe Emulation, in which the host is presented with an NVMe PCIe device, while the implementation of the device logic is done in software on the IPU's processors.NVMe emulation has many interesting use cases: bare-metal and virtual cloud providers; storage systems with proprietary protocols; full control, security and isolation over storage services; added features like compression and encryption over the standard NVMe interface; and more. The presentation will show NVMe emulation APIs for such IPUs, the use of SPDK as a standard framework for development, and offload possibilities.

Learning outcomes

1. NVMe emulation 

2. SPDK framework

3. RDMA

 
 

Track B Abstracts

What every technologist should know about AI and deep learning

Alex McDonald

Abstract

AI and Deep learning is disrupting every industry. For decades, AI was all about improving algorithms; and now the focus is on putting AI to practical use. This presentation will take you on a brief tour of Deep Learning systems; help you appreciate the role well-engineered systems play in AI disruption and making AI practical; and the impact on future of systems of applied AI in systems software design.


SNIA Long term retention in Medical AI Applications

Simona Rabinovici-Cohen

Abstract

In 2019, the SNIA Self-contained Information Retention Format (SIRF) became an ISO/IEC 23681:2019 standard. SIRF defines a storage container for long term retention that will enable future applications to interpret stored data regardless of the application that originally produced it. It can be beneficial in domains that need to keep data for long periods and enable search, access, and analytics on that data in the far future. The standard includes also examples of SIRF serialization on the cloud and on tapes. Radiomics is an emerging area in the medical AI community aiming to extract features from the patient multi modal medical images to improve the prediction of response to medical conditions. One challenge of radiomics is providing storage systems to efficiently store and preserve the patient medical data for future AI during the patient lifetime and beyond. The European Union H2020 BigMedilytics project includes a radiomics application that aims to predict response to breast cancer treatment.  In this talk, we’ll describe SIRF and how it can be applied in the BigMedilytics breast cancer pilot.

Learning outcomes

1. Recognize the challenges and value in the long-term preservation of big data for AI applications.

2. Learn about SIRF, its architecture and serialization for the cloud and for tapes.

3. Discuss the usage of SIRF for radiomics applications such as the BigMedilytics breast cancer pilot.


Samba for integrators

Volker Lendecke

Abstract

Topics for this session will centre around:

Samba architecture:

  • What components does Samba consist of?
  • What protocols does Samba implement?
  • What process model does Samba utilize?  

Samba SMB server:

  • What file system features does Samba expect?
  • Common complications and performance issues
  • Samba features for NFS interoperability

Samba authentication:

  • Overview of authentication mechanisms in the Active Directory world
  • The role of Samba's winbind daemon: Authentication and idmapping
  • Winbind API and services for non-Samba SMB servers

Samba Remote Procedure Call services: 

  • Requirements for seamless Windows interop beyond SMB
  • Protocol description for RPC services (direct TCP, named pipes)
  • Samba components that help non-Samba SMB servers

Computational Storage Workloads - Implications for Data Center Architectures

Jamon Bowen

Abstract

Xilinx is deeply involved in several computational storage products with a range of partners.  From this vantage point we have identified the common elements of workloads that map well to Computations Storage Drives (CSDs), Computational Storage Processors (CSPs), and Computational Storage Arrays (CSAs).  The properties of these workloads will be outlined with brief examples.  There is a growing effort to standardize the programming and deployment interfaces for computational storage and the programming interface (api) of Xilinx computational storage platforms will be covered with some discussion of extending computational storage over a fabric.

Learning outcomes

1. Understand what computational storage is and its benefits.

2. Understand the workloads for computational storage.

3. Learn about programing models for computational storage.


How to architect storage applications for the public cloud economy 

Josh Salomon

Abstract

The public cloud presents a new economic model on computing - "pay as you go" model in which you pay only for what you consume, buy you pay for everything: compute, storage, networking, QoS and more. This model suggests new considerations for applications architecture that minimize cost considerations, The presentation discusses why we need storage application on public cloud, the new architecture considerations (some of the discussion is applicable for all types of applications) and present several alternatives that can reduce the cost of storage applications in the cloud

Learning outcomes

1. Why we need storage applications in the public cloud (why not use the cloud provider services directly)

2. Understand the economic model of public cloud distributed applications

3. How we can architect and configure storage applications that will cost less when deployed on a public cloud.


Linux Access to the Cloud and NAS reborn: Recent progress in SMB3.1.1

Steve French

Abstract

Over the past year cifs.ko has been one of the most active file systems in Linux, with dramatic improvements to performance and new function which has enabled additional workloads to run over network mounts using SMB3 and SMB3.1.1.  This has made this a great year for Linux SMB3/SMB3.11 support. This presentation will demonstrate and describe some of the new features and progress over the past year in accessing Azure, Windows, Macs, and modern NAS servers and the cloud (Azure) via SMB3/SMB3.11 using Linux clients and  tools.  New configuration options to help performance, improve security and to enable additional workloads will also be discussed.

Learning outcomes

1. What new features are now available in Linux kernel and tools to access the Cloud, Windows, Samba, Macs and other NAS devices?

2. What new What SMB3 features should you expect in the next few releases of the Linux kernel?

3. Where has performance improved and what important performance features are still in progress?


Cloud-Native Standards-Based Clustered Storage with LightOS

Alex Shpiner

Abstract

Disaggregating storage from compute maximizes cloud builders' resource utilization, improves their flexibility, and enables them to move from an inefficient direct-attached SSD model to a shared storage model with independently scaled compute and storage. Lightbits’ LightOS is a software-defined disaggregated NVMe/TCP-based storage solution with increased performance, reduced latency, and useful data services. A cluster of LightOS servers replicates data internally and keeps it fully consistent, durable and available in the presence of failures. Data replication is transparent and server failover is seamless. Cloud builders do not need to install any client-side drivers as everything is done through standard NVMe/TCP. In this talk, we explain how to build cloud-native standards-based clustered storage with NVMe/TCP, introduce LightOS clustering, and discuss some of the challenges we encountered while building it.

Learning outcomes

1. Overview of clustered storage benefits.

2. Use cases and needs for clustered storage

3. Challenges of building a standards-based clustered storage solution for cloud environment


Accelerating the qualification of Solid State Storage

Leah Schoeb

Abstract

As the use of SSDs spreads throughout today's data centers, offices, and homes the importance of verifying their reliability and performance before deployment increases. Drive qualification requires the joint participation of suppliers and customers because datacenters lack the ability to monitor, analyze, and replay production workloads. On the other hand, SSD vendors cannot possibly understand the challenges and behavioral issues of real-world implementations without customer involvement. Meanwhile, both suppliers and customers want to accelerate the entire process as much as possible to reduce overhead costs and get drives into use as fast as possible. Architectures are currently available for accelerated validation, and a wide range of useful tools are available to support the effort.

Learning outcomes

1. Provide an overview of the difference in validating client storage to datacenter storage

2. Understanding the importance of storage power management in validation against standards

3. Technology differences in using client storage in an enterprise environment