2019 SDC India Abstracts

Plenary Abstracts

Fundamental Challenges to IOT Data Management

Vipin Shankar, Nitin Singhvi


Up until now, much of the apparent innovation of IoT has been centered around devices, platforms, ancillary technology areas like Edge, 5G, and the new use cases they create. Yet, the true value of IoT is in the data. IoT and its ecosystem is continuously and steadily creating new data silos making it imperative for enterprises to analyse and handle this data much more effectively than ever. IoT application areas such as smart cities, smart retail, connected devices, smart homes and mobility are demanding changes in the way this voluminous data is analysed. 

With data also comes the security, storage, governance, privacy aspects, and the need to realize its full potential. Furthermore, IoT being a mesh of devices, machinery, firmware and nodes, its critical to  ensure that the data is adequately transferred, stored and processed. This keynote will touch upon the fundamental challenges to IoT data management, and the nuances of Storage and New age Datacenter technology to support the ongoing IoT revolution.

Modern Data Protection Challenges

Kalyan C Gunda


This presentation will provide insights on how the data explosion has changed the protection landscape with IOT devices, cloud, data staying for longer times, and the challenges these bring out with edge devices, data centers etc.


The Rise of Computational Storage

Burzin Engineer


Driven by the demand for fast response time to users when challenged with huge data sets, Computational Storage has emerged as a fast growing data center infrastructure tool, especially with the transition to PCIe Flash.  Adding intelligence direct to Flash storage devices is a simple idea that can save expensive and time consuming data movement to the host CPU while parallelizing compute across storage drives.  This presentation will discuss the technology and market demands, application use cases and standardization efforts that are propelling Computational Storage to be a leading Flash based solution for modern database, big data/analytics, CDNs and emerging AI/ML workloads.


Discover PCIe Trouble-shooting

Manoj Vaddineni


  • Review the key elements within PCIe Gen4 specification
  • key challenges seen with the devices in the ecosystem at PCIe Gen4
  • How to effectively use error injection to validate the resiliency of the DUT
  • The best practices and approaches to testing at Gen4 speeds
  • The Road to PCIe Gen5 and the early eco-system

Intelligent Data Service Management

Ashit Kumar, Najmudheen CT


Enterprises use multi-vendor storage, multi-platform and multi cloud environments for IT workloads and data services. The metrics data from all these systems play a vital role in managing the data services and various run time actions for performance, reliability, availability and more. Usually SRM tools/like are used to manage this. However with the upsurge of data being collected and dynamics of the infrastructure maintenance, need of the hour is to make intelligent decisions based on the collected data; that too runtime! This is challenging. However, autonomous data management platforms like OpenSDS along with AI/ML can provide solutions. OpenSDS design can support integrations with SNIA Swordfish and similar specifications to accomplish better standardized data management platform. The session will cover the need of metric data analysis, advantages, challenges and possible solutions for efficient data service management. It will also cover how these data can be used efficiently for data service management like migration, anomaly detection, storage tier definition, data life cycle management, multi cloud management etc.

Learning Outcomes

a. Understand need of metric data analysis, advantages, challenges and possible solutions for efficient data service management. 

b. Understand solutions through open source data management tools like Open SDS.


Quantum network of secrets from India 

MT Karunakaran


With Quantum technology just around the corner, the existing cryptographic system will fall like a pack of cards. Securing data is essential for all organizations. Hence it becomes compulsory for us to take proactive steps today to secure against the future quantum adversary. At QNu we provide a quantum envelope to data, both in transit and storage in the form of Quantum Key Distribution and post-quantum solution. Currently, QNu Labs is in process of providing solutions to Indian Defense.





Experiences with DApp on Blockchain as a service  

Girish Chandrashekhar


This presentation is about development of DApp (Decentralized Application) on top of blockchain as a service. One such Application could be secure, shared ledger for data management operations.

Learning Outcomes

a. Why and where Blockchain is needed?

b. How DApp can be developed on Blockchain?

c. Current Limitations of Blockchain


Recommendation Engine for Data Eviction Policy in Cloud Service Providers  

Divyank Shukla


Quality of Service (QoS) in IO controllers is a mechanism for Cloud Service Providers (CSP) to conform to latency discipline for the IO requests received from its clients. Statically, there are multiple ways to associate or allocate hardware resources to service requests. One key challenge that the CSPs face is associating components with the IO requests on the I-T-L nexus of the client. The problem gets compounded by servicing backend IOs through a multitude of devices, with IO characteristics as high as milliseconds and as low as tens of microseconds.
An IO request or an Object Data Store configuration may generate additional auxiliary IOs to the system components. For example, we have seen RPO and RTO requirement drive IO latency expectations for clients.
We propose a neural network layer, that can classify IO requests based on IO patterns and enable data placement strategies/policies on the data stores. This can preemptively help storage controllers achieve its QoS objective and drive down costs.

Learning Outcomes

a. Dependable parameters for Neural network analysis for IO profiling

b. Tuning Hyper Parameters to improve classification

c. Tuning and or calibrating IO staging strategies in Storage Controller


Highly Scalable Cognitive Storage Management Platform using Cloud Native Services   

Ramakrishna Vadla, Maneesh Rapelly


In today’s competitive business environment, storage management providers are continually working towards improving the business value to their customers. The enterprises are deploying large scale distributed storage subsystems to cater high workload demands. The challenge for the storage management services is proactively finding performance bottlenecks, health checks, notify risks and prevent before they occur. The prior knowledge of the known issues from other customer storage deployments for correlation is a challenge. To address these challenges, they are implementing solutions based on the Artificial Intelligence including Machine Learning (& Deep Learning) which require large data sets to derive insights and run predictive analytics. The data from different customer deployments will give even better predictive insights. It requires architectural changes to the storage management services to deploy on the cloud designed by cloud native services those are reliable and auto scalable. The cloud native services such as kubernetes cluster, docker, lambda functions, object storage, elastic search, API gateway services and NoSQL e.g. dynamodb/cassandra for data lake are helping to manage storage infrastructure seamlessly. The cloud based AI services such as Amazon Sagemaker/ IBM Watson/MS Azure ML are used for integrating with data lake to run predictive analytics. The experience of addressing different storage management challenges using cloud native services will be shared.

Learning Outcomes

a. Inevitable need of Artificial Intelligence (AI) application in storage management

b. Understanding and choosing the right set of cloud native services

c. Future of storage management service architecture


India Cloud Storage Standards and Interoperability

Dr Dinkar Sitaram


Standards prevent vendor lock-in and allow the simple development of cross-cloud applications that will revolutionize computing in the same way as the development of the Internet. Under the charter of DoT and Meity, the Cloud Computing Innovation Council of India is working with TSDSI (Telecommunications Standards Development Society, India) to develop cloud standards for India that will be useful in the Indian context. This effort is slated to be complete by March 2020. The talk will give an overview of the methodology used as well as the current state of the storage standards.


Cloud Security: Current challenges and possible solutions

Anupam Jagdish Chomal


Many organizations, big and small, are moving their data into the cloud. New startups, (even banks) have adopted the model of starting off by keeping their entire data in the cloud. However, many do not properly assess its security implications. In this presentation, I will start by explaining different cloud architectures and the kind of security they offer. I will then explain the new attack vectors that have come up due to cloud storage and computing. I will give examples of how attackers have targeted cloud users and their data in the cloud. In the end, I will give my checklist on how data should be stored securely in the cloud.

Learning Outcomes

a. Understand various cloud architectures and compare their level of security offered

b. Learn about recent attacks on Cloud infrastructure and how they could have been avoided

c. Get a checklist of how data can be stored securely in the cloud


Storage Tiering with Deduplication

Nalini Nallamalli, Santosh Kalekar

Storage tiering is a feature that dynamically moves data between different types of storage to meet space, performance and cost requirements. Storage tiering policies place the most frequently accessed data onto the highest performance storage and less frequently accessed or old age data onto low performant or cheaper storage that at any given point of time one copy of data is present across all the storage tiers.

In case of environments where deduplication is enabled, same data chunk might get referenced by the multiple files/backups which may be cold or hot. So, it is not correct to directly apply tiering policy on such colder files and move all the blocks corresponding to cold files to low cost tiers, as same blocks might be getting referred by files which are latest and hot. It is very important that correct data chunks are identified and moved to appropriate storage tier when tiering policies are enabled. Otherwise overall performance will get impacted. 
This presentation talks about a solution on how to identify the correct data chunks to move across tiers when deduplication is enabled.

Learning Outcomes

a. Introduction of storage tiering

b. Introduction of how deduplication works

c. Challenges with Storage tiering in dedupe environments


Gen-Z envisages next-generation of memory management

Parmeshwr Prasad


In Gen-Z any component can be requester/responder based on capability and need. To enable application-transparent access to a Responder’s addressable resources, a requester maps each responder’s addressable resources into the requester’s memory address space. Memory pages are mapped to a requester page through the requester’s memory management unit (MMU) or logic that supports equivalent operational semantics. When an application on the Requester allocates memory, the allocated memory is mapped to a series of pages with some pages being mapped to a given Responder’s memory page. Once resources are allocated and mapped, an application uses load-store / read-write operations that are transparently translated using the Requester’s MMU and translation logic into Gen-Z read and write request packets to access responder resources. This provides efficient address access between components.

Learning Outcomes

a. Understand GenZ technology

b. Understand data transfer between different Gen-Z components

c. Benefit of GenZ to ecosystem and next generation of computing


Workshop on SNIA Swordfish Tools

Nidhi Malhotra


Workshop on SNIA Swordfish basic web client and Swordfish API emulator.

Learning Outcomes

a. Installation, usage and contribution to Swordfish basic web client

b. Installation, usage and contribution to Swordfish API emulator

c. Other SNIA swordfish tools


Optimizing data centers for the IoT Revolution

Geetkumar Chauhan, Kishor Jadhav


Internet of things has revolutionized the industry but it has given a big challenge to data center solutions. The rapid generation of massive amounts of data and it's processing requires data centers to be highly scalable and reliable. We will take a look at how this problem can be solved by enhancing the data center policies and storage tiering to optimize overall performance.

Learning Outcomes

a. Internet of Things

b. Data center problems due to massive amount of IoT data

c. Data center policies and storage tiering to optimize IoT data handling


Non Volatile Main Memory for Handheld Devices : An idea whose time has come

Manu Awasthi


Handheld devices, in the form of tablets, smartphones and wearables are becoming increasingly commonplace. Mobile applications with ever increasing feature sets are also being released. Owing to their rich features and large working sets, mobile applications are increasingly becoming constrained by memory capacity and bandwidth.
Non-volatile memory technologies like phase change memory (PCM) and magnetic random access memory (MRAM) provide attractive alternatives to LPDDR, owing to their superior density and energy characteristics. These technologies, in conjunction with LPDDR, can help design memory architectures with higher capacity and bandwidth characteristics, as well as better energy consumption profiles which is another primary design constraint for mobile SoCs.
However, extensive exploration of NVM technologies for memory hierarchies of mobile SoCs has not been done. This is in part due to a lack of modern tools and benchmarks required for such studies. In this talk, we describe some of our recent work to address these challenges. We first describe META, a memory hierarchy exploration tool for android devices, developed by our group. META allows for rapid exploration of NV memory hierarchies for mobile SoCs, across a number of Android versions. We also present initial, very promising results obtained for a few hybrid memory architectures designed using LPDDR and NVMs which allow us to demonstrate the feasibility of using NVMs for handheld devices.

Learning Outcomes

a. Memory design challenges in handheld devices

b. Tool design for memory architecture exploration

c. Potential uses of NV memory in handheld devices


Storage acceleration with CCIX

Dinakar Medavaram, Venkata Ravi Shankar Jonnalagadda


In this presentation, a brief introduction of new interconnect standard CCIX will be presented, followed by storage use case acceleration achievable with CCIX. CCIX is a new class of interconnect focused on emerging acceleration applications such as machine learning, network processing, storage off-load, in-memory data base and 4G/5G wireless technology. The standard allows processors based on different instruction set architectures to extend the benefits of cache coherent, peer processing to a number of acceleration devices including FPGAs, GPUs, network/storage adapters, intelligent networks and custom ASICs. FPGA acceleration has become a de-facto standard for obtaining the high computational throughput in present times. Storage domain can take advantage of FPGA acceleration for compute intensive tasks like compression, encryption etc. Through CCIX interconnect FPGA accelerators can act as coprocessors to host, sharing data structures seamlessly, with reduced data transfer latency. 

Presentation shall discuss two use cases 1. Memory expansion model over CCIX  2. Storage with compute offload. Details on programming model for realizing the true potential of CCIX interconnect will be presented.

Learning Outcomes

a. CCIX interconnect relevance & Introduction

b. Role of Coherent accelerators in compute intensive tasks

c. Programming model for realizing the true potential of CCIX interconnect


NFS Ganesha - Weather report

Jiffin Tony Thottan 


NFS-Ganesha is a user-mode file server for NFS (v3, 4.0, 4.1, 4.1 pNFS, 4.2) and for 9P from the Plan9 operating system. It can support all these protocols concurrently. The project was started around 2009 and it got well matured over past few years and includes participation including CEA, IBM, Red Hat. There are a lot protocol specific features added including LABELED NFS, Delegations to nfs-ganesha layer. The is workload specific changes made to nganesha layer which includes async op and non blocking io's. There are other projects like storhaug which integrates the nfs-ganesha to ctdb and so on.

The session will touch up briefly on all the new improvements happening in nfs-ganesha and how it can be consumed in different scenarios.

Learning Outcomes

a. NFS Protocols 4 and higher

b. NFS-Ganesha Project

c. Highly available NAS solution for NFS-ganesha


Track A Abstracts DAY Two


Multi-Cloud Paradigm: An Integration Approach

Pawan Ratwani


Evolution of cloud based technologies have revolutionized modern IT environment, while posing new challenges. Rapid generation of data, and variety of mechanisms to maximize its value have resulted in new cloud offerings rapidly. Since there can't be one solution to all the problems, its obvious that industry is inclined toward multi-cloud environment. In this talk we'll discuss about an approach to make the better use of multi-cloud environment.


NVMe-oF Ethernet SSD

Swati Chawdhary and Sandeep Kumar Ananthapalli


NVMe over Fabrics protocol (NVMe-oF) is designed to connect and scale NVM storage over a network with minimum latency overhead. However, existing NVMe-oF solutions cannot harness the low latency benefits of next-generation storage technologies, since these solutions utilize expensive CPU resources and require protocol conversion (NVMe-oF to NVMe) that adds overhead. This can be solved by developing an Ethernet SSD with native NVMe-oF protocol support. In this talk we examine the design of NVMe-oF based Ethernet SSD's and the use-cases they are enabling.

Learning Outcomes

a. Overview of NVMe-oF protocol and existing solutions

b. Design of NVMe-oF Ethernet SSD controller

c. Use cases


Emerging Ethernet standards and their impact on storage

Anupama BN


Over the past decade, Fibre Channel (FC) technology has been the norm when it comes to data centers. With storage applications such as data backup, replication and disaster recovery, it was used to connect storage to servers. FC backbone uses highest speed of 64 Gigabits per second (32G), which explicitly requires leased lines consisting of FC switches and bridges to be connected to Storage, and implicitly adding more cost to the infrastructure. Using latest technology of 100G RDMA (Remote Direct Memory Access), it is possible to achieve transfer speeds of 100 Gbps (Gigabits per second) with lower latency on the existing Ethernet switches and routers over long distances up to 300kms.  Will focus on how 100G Ethernet could replace the FC back-bone for storage connectivity with High Availability (HA) and Disaster Recovery (DR).Basically explains the front  (client-facing data ports ) and backend support (Storage and HA interconnect)  on 100G infra with Storage platforms .

This talk will mainly focus on NVMEoF  connectivity with RoCE v2 to achieve high speed storage access with low latency  and provide HA and Disaster Recovery seamlessly

Learning Outcomes

a. RoCE over Long distance

b. Cluster and HA on  Ethernet N/w

c. What is RoCEv2


Optane performance analysis on Android

Shyjumon Nankandiyil


The presentation showcases the performance and benefits of the Intel Optane storage as an SSD in Android OS. We have done comprehensive analysis of the Optane storage advantages and how it can impact on the Android OS workloads. Our analysis is majorly based on the Optane used as a SSD in a M.2. Interface. 
We made a fair comparison with the Samsung 960 Pro SSD (which is the best NVMe based popular SSD), we have analyzed the performance in various conditions such as boot, file transfer, Data base performance, AI capabilities and also on imaging and graphics workloads.

Also we are giving the opportunities which can bring in with the adoption of Optane in Android space in terms of new features. The analysis data is very exciting and the cost vs performance matrix is also interesting.

Learning Outcomes

a. Performance impacts in Android OS storage space

b. Use cases & advantages of Optane in Android/Storage space

c. NVMe storage capabilities and its future expansion possibilities


NVMe over fabrics - Demystified

Rob Davis


The new NVMe SSD interfaced can be connected across a Fabric. In fact it can be connected across lots of different fabrics: Ethernet (3 approaches), Fibre Channel, InfiniBand, and PCIe to date. Data Centers want to share storage readily among multiple compute nodes and be able to perform clustering, failover, and other system-wide operations at NVMe SSD speeds. NVMe over Fabrics (NVMe-oF) is the solution. This talk will describe the technology in its many forms. Describe use cases, for both Enterprise and cloud, where it is being applied. Then finish with potential future directions it is heading.

Learning Outcomes

a. NVMe over Fabrics

b. Multiple versions of NVMe over Fabrics

c. Future NVMe over Fabrics direction


Comparative performance analysis of NVMeFC and SCSI FCP

Mohit Chitlange, Saurabh Singh


SCSI is conventional block storage protocol built for legacy drives and is designed to cater wide variety of media ranging from tape to HDD drive. This introduced lot of serialization and complex IO stack in SCSI implementations. With introduction of fast storage like Flash drives which are capable of parallel IOs, SCSI protocol cannot derive full potential of the drive.
NVMe protocol has simple command set and supports multiple deep queues which can fully utilize flash storage performance by efficiently spreading the queues on multiple CPU cores.
This session details out performance comparison of SCSI-FCP Vs NVMeFC for variety of workloads. Precise benefits of NVMeFC due to efficient IO stack and inherent parallel architecture emerge out explicitly in terms of reduced latency, better IOPs and savings on CPU usage. The session clearly brings out how customer will benefit by low capital investment, addition of new workloads and ease of migration and administration.

Learning Outcomes

a. NVMe over Fibre Channel architecture and its advantage. NVMe over Fibre Channel internals and how it is one of the most appropriate host attach protocol for fast storage media.

b. Performance comparison of NVMeFC Vs SCSI over Fibre Channel. Comparative analysis of performance results on various metrics like response time, IOPs and server CPU utilization using various workloads.

c. Customer benefit on adoption of NVMe over Fibre Channel. Detailing out benefits like low capital investment, addition of new workloads and ease of migration and administration.


AI Based Ethernet Storage Management

Rishika Kedia, Dileep Dixith


The proliferation of applications for storing their data has increased adoption of Ethernet Storages operating over iSCSI/iSER/NVMe protocol. Its critical for business to maximize their availability implying the need to consistently monitor the performance of these storages to improve their efficiency. But debugging performance problems for Ethernet storages has below challenges
1.Instead of monitoring the storage in isolation, there should be broader IP monitoring including other infrastructural devices like ports, Switches and host ports. 
2.Predictive analysis considering whole of Ethernet storage fabric can proactively help identify the issues before their occurrence, this is because a configuration change in one component may impact performance of another component.
We provide the solution for Ethernet Storage management
•Meta-data collector agent at the client data center: A small footprint agent, installed on the client data center that,at regular intervals gathers performance metrics for storage that it is monitoring. For further processing uploads metrics to cloud storage management service 
•Cloud Storage management software: The Cloud-hosted SaaS service processes the uploaded from the agent and stores them to performs analytics on them. We are looking at applying data-science to understand the current status of environment and for any alarming situations it publishes alerts to the client, admins and support teams for further actions.

Learning Outcomes

a. Move a step ahead from single dimensional analysis of storage performance management from considering just stored-time-series data to multi-dimensional analysis

b. Understand how data-science is helping evolve storage management

c. Understand the challenges of Ethernet storage management domain



Applications have changed and someone forgot to tell storage

Abhishek Jain, Yadavendra Yadav


With DevOps becoming one of the most widely-used buzzwords in the industry, automated infrastructure management tools like Docker, kubernetes, Ansible etc. have been rapidly increasing in popularity and adoption. Persistent Storage for containerized stateful microservices such as MongoDB and PostgreSQL is one of its motive to drive storage vendors to plug in solution available to popular container orchestrator like Kubernetes. As containerization technologies enter enterprise market, they meet new functional demands such as providing and managing persistent, highly-available, parallel data access, yet a nimble clustered file system.
The proposed presentation is aimed to discuss in detail different challenges in designing a reliable  solution using parallel filesystem in containerized environment (showcasing Kubernetes, Cluster Filesystem as an example) and container lifecycle techniques to manage Persistent Storage using parallel filesystem.
Learning Outcomes

a. Understand how the clustered file systems are falling short for container workloads

b. Introduction to kubernetes's various approach in consuming Persistent storage

c. Learn pro's & con's using which a parallel, clustered filesystem in containerized environment


Container Storage Interface for Kubernetes

Anil Degwekar


The Container Storage Interface (CSI) is a standard developed by the Cloud Native Computing Foundation (CNCF) for accessing storage in any container environment. Storage vendors can develop a CSI driver for their storage systems once, and then these drivers can be used with any Container Orchestrator.  Kubernetes is a leading container orchestrator which has adopted the latest CSI spec (1.0).
GoCSI is an open source framework which provides essential tools for developing a CSI driver in the Go language.
As a follow up to my last year’s talk at the SNIA SDC India (‘Persistent Storage for Containers’), this talk will cover a detailed overview of the CSI standard and the GoCSI framework, and will provide insights on how to develop and test a CSI driver for Kubernetes.

Learning Outcomes

a. Understand CSI in Kubernetes

b. Understand GoCSI framework

c. Learn how to develop CSI drivers for Kubernetes


Blockchain and Storage - Bridging the gap

Karthik Rangamani


Blockchain technology has been growing in the last couple of years. Bitcoin has been synonymous with blokcchain and rightly so , as it has unlocked the potential of the Distributed Ledger Technology. Over the years there has been a sharp increase in the Blockchain space and with it a genesis of a new perspective has emerged. In this presentation , I would be talking about the impact Blockchain and Storage on each other and also look at some of the transforming business value that they bring together.

Learning Outcomes

a. Blockchain concepts

b. Business promises of Blockchain and Storage together

c. Blockchain use cases for the storage domain


Understanding the reliability of predictions made by Machine Learning

Rahul Vishwakarma, Supriya Kannery


Reliable estimation of prediction confidence remains a significant challenge in machine learning. We usually expect past performance to indicate future performance. When we deal with risk-sensitive systems – where the cost of a bad decision can be very high, and prediction accuracy is not the only objective; we need a multidimensional perspective about the forecast models. So, if the user is given a confidence of each new predictions made by the model, then a more meaningful action can be taken.

In this talk we will discuss about Conformal Prediction Framework and how it can be leveraged across various machine learning algorithms used in Storage Industry (e.g. Disk Drive Failure detection and storage demand forecasting). Furthermore, as an example, we will describe how this framework can be translated to time-series, classification and regression problem which will give a confidence (indication of the quality of each prediction) and credibility (filter mechanism with which we can “reject” certain predictions).

Learning Outcomes

a. Conformal Prediction framework for reliable time-series forecasting and application in regression and classification problems

b. Fundamentals of On-line learning approach and handling concept drift


How IoT, Analytics and ML unfolds in storage fabric

Sharath Thalya Shankarananda


A comprehensive presentation on how IoT, Analytics and Machine Learning can be applied in the Storage Fabric of a Data center. The presentation speaks about variety of sources (data end points), how data can be collected from the sources, how to store and understand the data, how to perform analytics on the data and provide insightful visualization, how Machine Learning can be applied for predictive analysis on multiple use cases and finally, how it will improve the performance, cost saving, customer satisfaction.

Learning Outcomes

a. A general direction on tools and application available to achieve results.

b. Gain insightful information on how analytics can be performed.

c. Datacenter/Storage Admin, C-level management and other stakeholders can take key decisions based on actionable information obtained.

Application Consistent Backup for Containerized Workloads

Jacob John Andathethu


This presentation discusses how  application consistent backup and restore of containerized workloads hosted on a Container orchestrator platform can be realized. This presentation includes how the quiescing and unquiescing of containerized workloads can be achieved to take application consistent backup and how data can be restored and recovered.

Learning Outcomes

a. What are the main challenges running workload on containers?

b. How do we solve the problem of dynamic requirement of persistent storage for containers?

c. How do we take application consistent backup of different workloads in Container orchestrator platform?


Back to Top