Data Value at the Edge

The Edge, while frequently discussed as something new, is in fact another technical turn of the crank. Fueled by an abundance of smart devices and IoT sensors, worldwide data creation has been growing exponentially, driving our customers and partners to innovate. For example, between 2016 and 2018, there was an 878% growth in healthcare and life science data resulting in over 8 petabytes of data managed by providers per annum. Dell Technologies has been at the forefront of this data revolution enabling our customers and partners to leverage these new sources of data to drive business. The process of data creation, transformation and consumption has taken on new meaning as devices have become more integrated in our everyday lives. How this data lifecycle adds value to our customers and partners is the subject of our post today.

Data Creation

“Data is fuel.” We’ve heard this spoken time and time again. While that’s true – it doesn’t convey the process the data undergoes, before it become something useful. “Data is fuel” is the net result of this process, not the genesis.

So, how do we get to this final, consumptive state with data? Data Creation is a constantly evolving mechanism driven by innovation, both in technology as well as in society. For example, the idea of remote patient monitoring has evolved, enabled by complementary technologies like 5G networks and IoT sensors. The ability for health care providers to securely retrieve data from smart watches, pacemakers, blood pressure cuffs, temperature sensors, electrocardiograms and insulin pumps (to name just a few) has driven a new paradigm of patient care and engagement. This wouldn’t have been possible a few decades in the past and, due to innovative approaches in networks, data management, and sensors, it represents one of many unique applications of the data creation process. Once this data is created, however, it must be transformed to be useful.

Data Transformation

Using the example of remote patient monitoring, the data generated by various sensors is unique. It has no intrinsic value as a “raw” data stream. Binary bits of encoded data provide no context, no perspective on what is happening with a patient. To fully understand, contextualize and derive useful consumptive value, it must be transformed. This transformation process extracts information, correlates and curates it through applications like artificial intelligence and analytics and provides it back in a human and machine-readable format. 1’s and 0’s become more than their sum and now, as transformed data, they’re ready to be consumed. Continuing with the patient monitoring example, the doctor receiving this information is then able to correlate and analyze these data feeds from a variety of sensors and sources and view them with an eye toward application. Recently, a Dell Technologies customer was able to increase their analyst-to-support staff ratio by greater than 100:1, enabling them leverage this data transformation to achieve better performance. As we’ve now seen, data is now one step closer to being fuel.

Data Consumption

By now, data has been created in various modalities, transformed by analytics and artificial intelligence and is ready to be consumed. Consuming data is more than just visualizing an output; it is the action. Our doctor has received remote patient data, securely viewed the correlated results and is now ready to provide diagnosis. The diagnosis is the net result of this generative (reproductive) model. Rather than being static or one-time-use, data consumption has taken on new meaning. Broadening this example, doctors use data to predict how to better counteract and treat disease. Machine learning models consume training data to learn to take future action and to create and transform the outputs into new capabilities. Manufacturers view vehicle data in extended reality (XR), peeling apart systems to be able to experience the real-time interactions between components. This generative cycle continues to evolve as technology advances, making the most of data’s kinetic energy.

Continue reading

The Evolution of the Data Warehouse in the Big Data Era

About 20 years ago, I started my journey into data warehousing and business analytics. Over all these years, it’s been interesting to see the evolution of big data and data warehousing, driven by the rise of artificial intelligence and widespread adoption of Hadoop.

When I started in this work, the main business challenge was how to handle the explosion of data with ever-growing data sets and, most importantly, how to gain business intelligence in as close to real time as possible. The effort to solve these business challenges led the way for a ground-breaking architecture called Massively Parallel Processing (MPP), or data sharing across multiple commodity servers with Direct Attached Storage (DAS), which kept data close to processing power. This was a move away from traditional symmetric parallel processing, where data was stored in centralized storage and accessed over networks. This worked well – for a while!

Silos

The imminent “explosion of data” present in all analytics/AI discussions is the driver for the evolution of data warehousing and business analytics. The success of the old architecture is also the reason why it had to change. You must reshuffle data quite often to keep data close to compute (whilst you run queries), while increasing data size as well as concurrent applications and users’ access to the data.

This was often resolved by creating data silos, standing up a new infrastructure for new applications, as well as performance tuning. This created a multitude of business and operational issues. While you often deal with the same data sets, they are extracted, transformed, loaded (ETL) and maintained differently across various applications, which means you have different analytical windows into the data depending on which application you are using. So which data is the source of truth? From an operational point of view, the cost to maintain various copies of the data, secure it, and maintain SLAs becomes another challenge.

The figure below shows a typical analytics environment highlighting the complexity and data duplication.

Data required

The deluge of data mentioned above is almost entirely caused by IoT and edge devices, plus an explosion in individual devices and users of social media. This is not your standard canned, transactional, and nicely-formatted SQL data. Instead, it’s usually semi-unstructured data that can sometimes only be accessed by utilizing a schema-on-read approach — which wouldn’t work for traditional SQL Data Warehouses. Also, new data types are enabling (and, perhaps even forcing) the hands of businesses to look beyond a traditional analytics approach. Today, Hadoop is a large part of many organizations’ analytical capabilities, and AI is rapidly taking off, giving businesses 360 degree analytics capabilities, from descriptive all the way to prescriptive analytics. Traditional architectures wouldn’t be able to do this, at least not at scale.

Now what do we do?

IT infrastructure must evolve to meet these data demands. Modern analytics and business intelligence now encompass all aspects of business with a much larger, more complex and multi-faceted data set and application portfolio. How do we manage all the data and applications? Simple, the answer is a “true” data lake. Account for eliminating data silos, create a single accessible data source for all applications, and eliminate data movement/replication, so organizations can control compute and storage ratios to achieve better cost with the ability to leverage best-in-class elastic cloud storage and maintain open data formats without lock-in associated with proprietary data management solutions.

At Dell Technologies, we’re working closely with some of the leading software vendors in the data lake and data warehousing arena such as Dremio, YellowBrick, and Vertica to enable our customers to seamlessly build and deploy a scalable and future-proof IT infrastructure. Dell EMC Isilon and ECS provide best-of-breed and market-leading architecture that enables IT departments to use scale-out NAS and object storage technology that natively supports multiple protocols such as NFS, HDFS, SMB and S3. And all on the same hybrid file system offering cost effective performance, scaling and a complete set of enterprise governance features such as replication, security, backup and recovery.

The proposed architecture enables IT to eliminate multiple landing zones and ETL zones and the need to replicate data. Simply load the data files into the data lake and point your application to query data directly from the data lake. It’s that simple!

As an example, the figure below shows an open data lake architecture incorporating a data lake engine from Dremio and data lake storage via Dell EMC Isilon and ECS. Dremio delivers lightning-fast queries directly on ECS, offering a self-service semantic layer enabling data engineers to apply security and business meaning, while also enabling analysts and data scientists to explore data and derive new virtual datasets. Dremio ensures flexibility and openness and lets you avoid vendor lock-in, directly query data across clouds or on-prem, optionally join data with existing external databases and data warehouses, and keep your data in storage that you own and control.

Dell Technologies will continue partnering with key ISVs such as Dremio and Yellowbrick, to deliver AI and Data Analytics solutions that will allow our customers to power their journey to AI.

Continue reading

Dell Technologies and Ververica: Analyzing Continuous Data Streams Across Industries

Every single entity in the digital world, be it an end user or a sensor device, continuously produces activity updates. People leave traces of their activity when they shop online or use any other online application. Machines then report on actions and statuses. Capturing such traces and reports in the form of data streams enables software applications that  digitally reconstruct the history of these entities to allow queries against the data and actionable insights. The same is true for applications that represent these streams in different formats.

To satisfy the demanding requirements of such applications, we developed the Dell EMC Streaming Data Platform (SDP). The solution ingests, stores and processes data continuously, mapping the software abstractions more naturally to those types of applications. The centerpiece of our SDP platform is the open source storage system called Pravega.

While ingesting and storing are critical functions of any data pipeline, a streaming data solution needs a stream processor that can benefit from the features that Pravega offers. Pravega exposes streams as storage primitive, enabling applications to ingest and consume data in a stream form. Pravega streams accommodate an unbounded amount of data while being both elastic and consistent.

Apache Flink, a unique framework in the space of data analytics, offers powerful functionality for processing both unbounded and bounded data sets. When used in conjunction with Pravega, Apache Flink can tail, or historically process, data using the same source, while providing end-to-end, exactly once semantics, and dynamically adapt to resource demands. The combination of Pravega and Apache Flink in the Streaming Data Platform raises the bar for stream processing platforms. It brings an unprecedented level of features that have long been needed to fulfill the requirements of existing and future applications.

Figure 1: Data pipelines with Pravega and Flink in SDP

The Future

Dell Technologies envisions a world in which data streams are ubiquitous and stream processing is the new norm for modern application development. Stream applications are already prevalent, but there’s a clear increase in demand and scale for systems that process stream data. We are excited about this future and the opportunity to build the systems that can help our customers’ applications process stream data efficiently and effectively.

The combination of Pravega and Flink in the Streaming Data Platform to compose data pipelines already provides unique possibilities. The partnership between Dell Technologies and Ververica will further enable us to build features for data pipelines that will help organizations simplify storage needs, create a foundation of unified data and innovate using an endless array of applications. We look forward to working together to deliver streaming data technology that continues to provide meaningful outcomes for our customers.

Source: dellemc.com

Continue reading

Chaos, Cacophony, and the New Data Order in the Coming Decade

The total volume of data transmitted over the US mobile phone network in 2007 amounted to 86 petabytes. Today, the same volume is transferred in less than a day. So what will things be like ten years from now? We might only need ten minutes to transfer this amount of data, which would correspond to a 52,000-fold acceleration. It is safe to assume that the global increase in data transmission volumes will be in the same ballpark. At Dell Technologies, we refer to the next ten years as the “next data decade” and have aligned our 2030 goals accordingly.

Dell has always set itself ambitious and long-term goals. Nearly ten years ago, for instance, we announced our “2020 Legacy of Good” plan, based on the idea that technology should be the driving force behind human progress. We adopted this key idea in the plan for the coming decade, which will, of course,lead us to a completely different world from the one we know today. We will generate vastly more data, because everything will be connected to everything else. However, data has no intrinsic value; we need to bring it to life to make it usable in areas such as industrial automation, which will continue to advance; in autonomous cars, which will shape the urban landscape by communicating with other vehicles, traffic lights and the power grid; in healthcare, which must be drastically improved worldwide; in smart homes, smart cities, smart government, in our personal well-being, and basically in every conceivable sphere of life.

Not just big data (the term should actually be renamed incredibly enormous data in the new decade), but also thousands, millions, or perhaps even more fragmented databases and storage spaces will shape everyday life. This isolated data is nothing but chaos and noise until it interfaces with the rest of the world. The integration of all these data silos into countless clouds and on-premises systems is the Herculean task that companies and the IT industry are facing. Only when we solve the question of data integration will new technologies such as artificial intelligence and machine learning blossom and play a truly fundamental role in our future society. After all, these systems need to make a vast number of sometimes vital decisions based on this data jumble in real time.

However, this will only be possible if data security is also ensured. Therefore, data integrity and encryption will play an equally important role in the coming years. There are formidable factors to consider, among them the development of the quantum computer, which will completely revolutionize the way we do IT while also providing cyber attackers with a tool to crack almost any asymmetric encryption in mere seconds. We must therefore work to develop effective quantum-safe methods. Fortunately, the US organization NIST is already developing new standards for post-quantum cryptography.

And there are more challenges in store for us. For example, we cannot simply allow AI, which is being made more powerful by the day, to act like “black boxes.” We must ensure that decisions remain comprehensible and algorithms transparent, if necessary based on legal standards. This is no small feat in today’s globalized world. We must also ensure that the human being remains the key focus of technological advancement. In other words, we need to uphold ethics and human values in a fully digitized society – digital humanism if you will. And we must increasingly protect our environment.

All of this is part of our Next Data Decade plan, in which we will focus on product sustainability above all. Our concept here is what we call one-to-one recycling: for every product a customer purchases, we will fully recycle or repurpose a similar product. Second, we will promote inclusion with the goal of having 50 percent of our global workforce made up of women by 2030. Third, we strive to improve lives on a broad scale based on technology, that is, for billions of people. Fourth, we want to do so while upholding ethics and data privacy.

Do we know whether we will achieve these goals? Do we have any idea how information technology, let alone our society, will evolve over the next decade? Of course not. Can we anticipate stumbling blocks? Certainly. Will we be able to handle big data, AI, data security, and technological innovation over the next ten years as envisioned today? Even though the future may be uncertain, we are optimistic and will do everything in our power to advance human progress by leveraging our technology, our knowledge, and our compassion. This new decade is certain to be an exciting one!

Source: dellemc.com

Continue reading

Time to Accelerate Your 5G Journey

This time last year, there were a slew of predictions around the launch of 5G and its potential impact on the world. I personally forecast that Dell Technologies would help define the underlying infrastructure technology, serving as the rocket fuel to power the “connected everything” revolution.

One year later, have the predictions come to pass? Has reality met our expectations? All signs indicate a resounding yes. More importantly, the story of transformation continues to unfold.

Torrents of data

We’ve already seen the pace of innovation accelerate dramatically with the combination of 5G, Edge and IoT applications driving literally torrents of data from the Edge to the Core to the Cloud. According to estimates for this year, every person on earth is generating 1.7MB of data every second. Incredibly, this will result in 40 times more bytes of data than there are stars in the observable universe. That’s a pretty mind-blowing statistic.

In a way, it’s not surprising since we’re increasingly surrounded by always-on, always-connected smart devices. However, as you can imagine, the speed and efficiency of processing this growing workload is only as good as the network and infrastructure it runs on. Therein lies the challenge.

Implications for network builders & service providers

For network builders and service providers, the immediate priority is to meet the current demand for more data. However, you have to think long-term and position your network for the future. To capitalize on the 5G opportunity, you need to deliver massive speed, scale and innovation. You have to automate and cut costs in order to compete and speed your time to market. In short, you have to design the telecom network of tomorrow, today.

A modern network for the future

So, what does this modern network look like? In my view, you need built-in intelligence coupled with a reliable and yet simple infrastructure that can scale to meet future demand. Must-haves include carrier-grade platforms, based on open standards and commercial off-the-shelf compute, industry-certified solutions, best-of-breed switching and storage solutions, all fully optimized and designed to deliver a richer, more intuitive user experience. Think software-defined, flexible, multi-cloud and virtualized solutions.

5G and edge computing – a symbiotic relationship

In this new world, 5G and Edge computing—working together in tandem—allow network operators to deliver new business insights, reduce the costs associated with backhauling data to the cloud, and avoid network congestion. Most importantly, this combination is allowing network operators to intersect with other vertical industries—like video surveillance and industrial automation—to support applications that rely on low-latency connections, such as autonomous vehicles, smart factories and automated warehousing.

For example, one of our OEM customers that builds global networks is currently using Dell Edge gateways, Virtual Edge Platform 4600 and VMware Pulse to deliver a turnkey solution that solves connectivity for the smart factory or warehouse, relying on LTE technology. My colleague Ron Pugh recently touched on some other exciting Edge opportunities for the telecom industry. I believe that this is just the beginning; we’re merely scratching the surface of what’s to come.

 A track record of delivery

In this fast-changing environment, Dell Technologies OEM | Embedded & Edge Solutions has delivered these solutions successfully in the IT market for years, making us uniquely qualified to help network builders and service providers in this fast-changing environment. With expert knowledge of the telecom market, we can deliver industry certifications, ruggedized products, branding, customization, BIOS modifications plus a tremendous amount of engineering capability to help you to build the right solution that can deliver a software-defined, flexible set of capabilities.

Customer examples always tell the story best. With its telecom customers wanting to deploy a virtual session border controller in their networks, Ribbon decided to source a turnkey NFV platform from Dell Technologies OEM | Embedded & Edge Solutions to serve as a high-performance, hosting infrastructure, capable of high packet throughput, low latency and scalable audio transcoding. You can read the full Ribbon story here.

Meanwhile, Mavenir offers a 5G core solution based on our infrastructure and VMware container platforms. And in the world of telecom network testing, VIAVI Solutions integrates our XL servers with its custom FPGA cards and Linux-based application suite.

Innovation will take its own path

I believe that the future for the telecom industry is brighter than ever. No doubt, there are some developers sitting in a garage somewhere right now saying, “I wish I could make this thing work…if I could only get ubiquitous, cost-effective, latency-free bandwidth!”

As 5G starts to become more mainstream, I believe that we’ll experience tons of new innovation, new capabilities, new ways for people to connect with one other, new ways for businesses to connect. As a result, we’ll see new revenue streams, new business models and new markets emerging. Change is going to open brand new opportunities that none of us have even dreamed of yet. That’s the bit I’m most excited about, as innovation takes its own path.

Continue reading

Dell EMC Streaming Data Platform Enables Insights for Streaming Data from the Edge

Streaming Data Creates Massive Potential for Organizations Across Industries

According to IDC, more than a quarter of data created in the global datasphere will be real-time in nature by 2025. Much of that data will come from the edge, originate from sensors, cameras and drones, and come in the form of a continuous data stream.

Streaming data creates additional complexities in the already intricate world of unstructured data. With its tendency to vary in volume and boundaries and with timestamps that can fluctuate out of order, the need for a Data First infrastructure – where organizations know exactly where their data is and how to harness its full potential – becomes more critical than ever. While this type of data has the potential to drive significant innovation within an organization, it also presents those that work closely with enterprise infrastructures with a host of challenges. These challenges can include managing multiple infrastructures, dealing with data inconsistency risks and creating a time-consuming process for application developers.

Imagine providing your developers and operators with a platform purpose-built for them to ingest, analyze, and manage streaming data. Imagine monitoring anomalies in a manufacturing line, an energy field or railroad tracks through sensors – abnormalities of livestock, airplanes and shipping containers tracked through drone video feed – or the ability to adjust traffic lights depending on volume or detect security threats through video streams. These are just a few of the possibilities when an organization creates a Data First infrastructure focused on a proper streaming data foundation.

Capture the Streaming Opportunity with Dell EMC Streaming Data Platform

To help customers turn this unique set of infrastructure challenges into opportunities for deeper business insights, Dell Technologies is introducing the Dell EMC Streaming Data Platform. This all new software offering enables ingestion and analytics of real-time streaming edge data in the data center.

With the release of this enterprise-ready platform, organizations now have access to a single solution for all their data (whether streaming or not) that provides auto-scaling ingestion, tiered storage with historical recall on-demand and unified analytics for both real-time and historical business insights.

Re-Designing Complicated Infrastructures with an Out-of-the-Box Solution

As many organizations have begun to address the rise in streaming data by creating additional infrastructure solutions, they have inadvertently created duplicate data silos along the way that each require individual implementation, management, security and analysis. The Streaming Data Platform unifies those infrastructures into one out-of-the-box solution to streamline data and its management for actionable insights.

The software platform provides enterprise-grade security and serviceability and is built on open-source technologies such as Kubernetes for orchestration, Apache Flink for real-time and historical query processing and Pravega for ingestion and publish-subscribe. The Streaming Data Platform empowers innovation far into the future – creating a programming model that reduces application development time, giving your team more time to focus on innovation and the next level of business needs.

Improving Workflows and Outcomes Enterprise-Wide

As a platform that enables insights to drive innovation, the Streaming Data Platform enables a wealth of opportunities to endless industries – from retail, to agriculture, to automotive and the energy sector. Example uses of this technology include the real-time & historical analysis of video and telemetry streams to monitor livestock health or manage planning of large construction projects. One customer, RWTH Aachen University in Aachen, Germany, creates research projects for nearby small businesses – allowing them to reap the rewards of research that creates more efficiency, deeper insights and frees up time to enable innovation in use cases like the manufacturing example above. “The reduction of management needed to operate the software system and [Streaming Data Platform] stack is crucial … to ease the work of developers and minimize hard DevOps skill requirements,” said Philipp Niemietz, Head of Digital Technologies Research Group at RWTH.

Addressing the Streaming Data Challenge

When one thinks of enterprise storage, file and object are likely the first categories that come to mind. But with the rise of real-time data, it’s clear that in the years to come, storage solutions for streaming data will become a critical need in the enterprise.

Continue reading

Think You Need Different Systems for AI, Data Analytics and HPC? Think Again

With the new Dell EMC HPC Ready Architecture for Artificial Intelligence and Data Analytics, you can now run AI, data analytics and HPC workloads on a single system.

In today’s marketplaces, competitive advantage increasingly goes to the data‑driven enterprise. This is particularly true for those enterprises that are ready to seize the day — and capitalize on the convergence of artificial intelligence, data analytics and high performance computing.

The time is ripe for this convergence, thanks to some amazing advances in HPC systems and the applications that can run on them. As HPC clusters become smaller, simpler, and less costly to deploy and operate, enterprise IT teams can use HPC to provide the throughput and capacity they need to accelerate AI, data analytics and other compute-hungry enterprise workloads, like modeling and simulation.

If your organization is on this path, you’re headed in a great direction. When you converge AI, data analytics and HPC on a single system, you gain the horsepower you need to run high‑performance data analytics, simulation, boost high‑frequency trading, help with drug trials, enhance risk analysis, improve fraud detection, collect and analyze data from the Internet of Things, and accelerate motion picture animation and special effects cycles — just to name a few of the opportunities that come with convergence.

And this brings us to the news of the day — the launch of the new Dell EMC HPC Ready Architecture for AI and Data Analytics.

One system for AI, data analytics and HPC

If you’re in an IT shop, you know that system configuration can be a complex task, requiring a delicate balance among workload requirements, performance targets, data center constraints and your IT budget. With the new HPC Ready Architecture for AI and Data Analytics, you’re home free when it comes to these challenges. Engineers from Dell Technologies have done the heavy lifting for you, so you can confidently and quickly deploy one system for AI, data analytics and HPC workloads.

Like all the other Dell EMC HPC Ready Architectures, this new offering is designed by expert engineering teams to simplify system configuration, deployment and management. The HPC Ready Architecture for AI and Data Analytics has been optimized, tested and tuned for a variety of applications on the Kubernetes stack, with ongoing testing and validation to expand the list of validated options.

Flexible workload management enables dynamic movement of jobs between Slurm and Kubernetes based on user demand, with a scalable shared filesystem to support both. Bright Cluster Manager provides a single‑pane‑of‑glass management experience for Dell EMC hardware, Slurm and Kubernetes. The solution includes a set of Ansible and Terraform playbooks for those who like to build via open source tools.

This is a package, not a single point-product. The HPC Ready Architecture for AI and Data Analytics comes with a best-practices guide and toolkit to help you go from a factory‑installed operating system to a full Kubernetes cluster with a repository on GitHub. And with Bright Cluster Manager software, system admins can quickly get clusters up and running and keep them running reliably throughout their lifecycles.

Great ROI

There is even a return-on-investment (ROI) side of this story. A detailed total cost of ownership (TCO) analysis by Dell Technologies showed that the ability to run multiple workloads on a single infrastructure lowers the cost for hardware, software and maintenance, while providing additional savings from reduced power and cooling costs. This comparative analysis demonstrated that with a converged solution, system users can get faster access to more comprehensive datasets, while IT teams can reduce silos and consolidate operations for up to 3x lower TCO. With this new HPC Ready Architecture, you can now run AI, data analytics and traditional HPC workloads on the same system — simplifying deployment and management while keeping your costs low.

Continue reading

Dell Technologies and Splunk: Enabling AI and Analytics Together

I’m a Business Development lead with the Dell Technologies Unstructured Data Solutions (UDS) team. In this role I drive go-to-market and enablement for enterprise data center storage offerings with many of our partners in the Data Analytics and AI Solutions space. One of our key partners in this area is Splunk and I’ve been working with them for quite a while now. In fact, I recently attended .conf19 in Las Vegas with over 11,000 Splunkers and it opened my eyes. Let me tell you why I’m so excited about the possibilities for customers working together with Dell Technologies and Splunk on their journey to AI.

Splunk is already one of the largest data aggregators in the industry. I personally evaluated and deployed Splunk back in 2007 at a large media and entertainment company where we were using approximately 50,000 render cores with data coming in from numerous machine logs, applications, data bases, schedulers, render farms and many other sources.  With this level of data consolidation there’s a need to apply machine learning for predictive learning and root cause analysis, a significant use case for Splunk.

Data growth continues unabated at an exponential rate due to the number of connected devices talking to the network. Everything today has sensors, with all of them transmitting data to some central repository for analysis along with other related data sets. With these and other IOT devices being so prevalent and decentralized we need globally distributed storage repositories to be able to receive this data.

Last year, Splunk announced availability of a new modern data path model called SmartStore. SmartStore allows you to decouple compute from storage and also supports the S3 protocol. You still have the hot/cache tier as tier 0 but now you can tier from it directly to a SmartStore target like Dell EMC’s Elastic Cloud Storage (ECS) which has been certified by Splunk. In addition, Dell EMC also released an ECS Splunk plugin freely available for customers who have deployed ECS and want to monitor it via Splunk. There are links below for information and downloads.

In the lead-up to Splunk .conf19 our friends from Big Data Beard  set out on an epic roadtrip from Boston to Las Vegas driving an RV loaded with IOT sensors, collecting data and storing it on ECS via SmartStore. With the power of the combined offerings of Dell Technologies, Splunk and other sponsors, the Big Data Beard team captured data, spotted trends, predicted failures, and discovered more exciting ways to use Splunk to extract value from machine generated data. It was a great real time and real-world use case that generated a lot of excitement at the event.

Dell Technologies had three sessions at Splunk .conf19 that are archived at the links below.

◉ “Splunk Apps for Infrastructure from Dell EMC”. In this session, you’ll hear from an experienced Site Reliability Engineer how Dell EMC’s investment in Splunk apps across their platforms makes it easier for Splunk users to monitor infrastructure and integrate these insights into an overall application performance monitoring strategy.

◉ “Cloud-scale On-Prem: SmartStore Best Practices with Dell EMC”. In this session learn how Dell EMC delivers cloud-scale object storage for on-premises deployments of SmartStore and how you can leverage best practices learned from some of the world’s largest global Splunk deployments.

◉ “Secrets from a Splunk Ninja: Deployment Architecture Best Practices”. In this session learn more about the best practices that power some of the largest, on-premises deployments of Splunk around the world. Determine if SmartStore is right for your environment and size it for maximum performance and benefit.

Dell Technologies and Splunk offer a few ways to help customers tackle AI, ML, or DL challenges with prebuilt, pretested, preconfigured Ready Solutions or Reference Architectures, all with or without GPU’s.

As data growth continues the opportunities for Dell Technologies and Splunk customers expand. 

Source: dellemc.com

Continue reading

What separates AI-Enabled Data Capital Achievers?

I recently joined the Storage Solutions team at Dell Technologies, which is tasked with marketing for High Value Workloads and Vertical Solutions across our storage portfolio. Our team handles everything from SAP, SQL and Oracle to Healthcare, Life Sciences, Media & Entertainment, Video Surveillance and Advanced driver-assistance systems (ADAS) – just to highlight a few. All these established solutions areas have important use cases and workloads that drive significant amounts of storage capacity and revenue.

My area of focus is somewhat different and unique on the team: I cover AI/ML/DL as a discrete solution area but also as a horizontal enabling technology ingredient across all vertical solutions. One of my key responsibilities is to showcase how our storage portfolio enables the design and deployment of optimal IT environments for AI initiatives. This is an emerging area, quite dynamic and diverse, made all the more important as our customers look to gain competitive insights and revenue advantages in this Data Decade.

We’ve all heard that “data is the new oil” and therefore one of the most important assets for organizations around the world. But the advantages and importance of this Data Capital are only realized if value can be extracted successfully. In order to effectively harness the power of data, the very backbone of your IT infrastructure needs to be optimized to handle the work. This can be achieved in many ways; however, one thing is certain and urgent – the data keeps coming and it won’t stop. We’re living in a time of unprecedented growth of unstructured data. Organizations are swamped and they need new ways to address this data deluge.

So how do you effectively and efficiently extract the value from data? What are the key ingredients of a modern IT Infrastructure for AI? What’s holding you back? Both Dell EMC and Intel, as IT industry leaders, are tasked with helping our customers answer these important questions as we collectively embark on our journey toward a digital future. To this end, we recently commissioned Enterprise Strategy Group to survey organizations around the world to determine their state of AI readiness. The results are published in this new Research Insights Paper: How Organizations Unlock Their Data Capital with Artificial Intelligence. The analysts were determined to discover what sets organizations apart with respect to their use of AI technologies and the best IT Infrastructure to power these demanding workflows.

A key tenet of the report is that legacy data analytics techniques simply will not work today and that forward-thinking organizations must implement AI into the mix in order to remain competitive. AI has significant advantages compared to traditional analytics, offering speed, scale, impartiality, precision and uptime. The report goes into detail on these facets, as well as the positive revenue and competitive impacts organizations can expect by introducing AI deployed on a modern IT infrastructure, summarized in the graphic below.

Although the results vary across the study’s respondents, it’s clear that no matter where they are in the evolutionary process of applying AI to their data analytics initiatives, the effort will drive significant amounts of incremental revenue.

How does your organization do the same? This highly informative ESG report highlights four key characteristics of Data Capital achievers:

1. Deployment of high performance and scalable storage.

2. Use of hardware accelerated massively parallel compute for optimal AI model training.

3. Adoption of comprehensive protection for the data pipeline.

4. Enlistment of third parties to augment data science initiatives.

It’s still early days in the use of AI to power data analytics initiatives. Even though, it’s clear from this report that organizations starting now to build out AI enabled modern IT Infrastructures are the ones that stand to gain the most from their nascent investments.

Are you interested in learning more? In addition to this ESG report, we’ve recently commissioned a couple other new research papers in this area. Please check out this report from Moor Insights & Strategy as well as this IDC Technology Spotlight.

Continue reading

Lemons or Le Mans: Why Not Using A Processor-Optimized System Will Leave You Underwhelmed

An Impractical Approach

Would you ever think that a smart car could beat a souped-up sports car on a quarter-mile race track? You could, by modifying the tiny smart car with an over-powered engine packed into its lightweight frame. This is a clever trick to get maximum power over a short distance. However, would you ever race one of these cars on an F1 track? Or tow a boat? Or take kids to swim practice?

Although these mental images are entertaining, a super-powered smart car does not make a useful or effective value prop for these activities. Think of the stress the engine would put on the brakes, chassis, and steering. Think of the maintenance, component upgrades, and labor that are required to operate such a car.

A Pragmatic Approach

Servers are designed and built in the same way – for specific workloads. They are not the sum of their individual components. Each piece of hardware must be optimized to work with other hardware and firmware to effectively tackle specific workloads. A powerful component without the right support does not perform at its full potential.

If you take the engine of a race car and install it into the frame of a midsized sedan there will be significant performance left on the table. This is exactly the case with dropping-in the 2nd Gen AMD EPYC processor (code named Rome) into a server that is designed for the 1st Gen AMD EPYC processor (code named Naples).

This makes one wonder about the release of AMD’s 2nd generation EPYC CPU. How will you effectively leverage this technology? Does a drop-in make technical or business sense, especially when comparing to a Rome-optimized system?

Technical Sense

If you’ve waited in line to check out at a retail supercenter – you experienced how the throughput of a system is dependent on the slowest part. This may induce anxiety when thinking about replacing old CPUs with new, advanced ones. By including a Rome processor on a Naples-based platform, you will experience lower performance, decreased capabilities, slower memory speeds, subpar networking, and limited platform scalability. Your memory and the input/output latency will slow your AMD 2nd generation EPYC CPU with 64 cores like a boy scout loaded with a fridge worth of food plus his family’s collection of kitchenware.

Business Sense

Using a server effectively also has business and financial implications. Cobbling systems can become labor intensive when a rack goes down because of an aged component. Operating costs can be 10 times higher in years 4 through 6 than the initial procurement cost of the server. Refreshing your servers around the three-year mark is shown to reduce overall costs. This is just the cost of operations. This does not account for the better outcomes and innovative solutions your employees will create, when they are free to pursue non-maintenance tasks.

Returning to the original analogy, one size does NOT fit all. The Toyota Prius could tow a boat. But why not use an appropriate car or truck? Matching the right workload with the right server will increase performance, automate management, and improve security (i.e., Dell EMC PowerEdge Servers with 2nd generation of AMD EPYC). This includes:

◉ More NVMe for better virtualization and software-defined solutions
◉ Increased cores per socket for hyper-converged infrastructure and virtual machines
◉ Lower latency, Gen 4 PCIe, with GPU slots for data analytics, artificial intelligence, and machine learning

Red Light, Green Light, Go!

Watching a smart car beat a Mustang can be entertaining, but is it a pragmatic solution for towing boats or everyday commuting? Should you drop-in a 2nd generation AMD EPYC chip into a Naples-based server? We all get excited when a new version of a technology we love is introduced. The hardest part is waiting! Dell EMC PowerEdge is releasing a portfolio of servers that are designed and optimized to leverage the full capabilities of the 2nd generation AMD EPYC processor.

Continue reading

Delivering a Modern Streaming Architecture for 5G

What is the Use for Data Analytics?

Analytics offers many benefits to organizations as they embark upon digital transformation, including:

◉ Increasing efficiency and driving cost out of operations
◉ Maximizing customer satisfaction
◉ Developing new products and services
◉ Use of streaming data to respond to issues and opportunities in near real-time

The number of use cases made possible by data analytics seems limitless and, on top of that, we are only now beginning to glimpse the potential of machine learning and other forms of artificial intelligence to open new frontiers of what organizations can achieve with data.

But as we at Dell Technologies engage with customers on a variety of use cases, the more we learn that many are still struggling with the prerequisite task of getting data into their analytics environments in order to deploy the use cases they want. This task is called ETL, or “extract, transform, load”, and it can be defined as the process of reading data from one or more data sources, transforming that data into a new format, and then either loading it into another repository (such as a data lake) or passing it to a program.

Dell Technologies and its telecom data analytics ISV partner, Cardinality have been working together to help customers resolve complex ETL issues so that they can do the kind of analytics they want. What follows are real-world examples that illustrate three key pain points customers tend to experience with ETL, and how we have helped resolve them.

Data Myopia

Telcos sit on a wealth of data, but organizational or technical barriers can often make it difficult for data engineers and data scientists to gain access to the data they need. The data analytics team at one tier-1 telecom operator faced just such a challenge. Only able to access data from the IT environment, the team couldn’t get the data they needed to start answering questions about the factors that influence customer satisfaction. To solve this problem, Cardinality conducted a pilot on a small footprint of Dell EMC PowerEdge servers to demonstrate to the Network Operations team the value that could be unlocked with a simple use case: device analytics. In a matter of days after configuring its ETL Engine to ingest data from the operator’s network probes, Cardinality was able to produce a real-time dashboard of all the mobile phones and other devices on the network, and show vital information such as types of SIM cards the devices where using and which could be upgraded to 4G networks. This operator was able to build on this initial use case to create a complex, Network Customer Experience use case that delivers measurable business benefits by using machine learning to analyze over 350 network KPIs in order to predict and circumvent customer churn.

Creeping Complexity

New technology spaces typically offer developers a wealth of tools to choose from. Many tools, both open source and proprietary, exist in the world of data analytics (e.g., Informatica, Talend, Kafka, StreamSets, Apache NiFi, Airflow, and many more). While choice can be good, the use of too many tools by too many different people in a single environment can make management a costly ordeal.

One telecom operator that Dell Technologies recently worked with had fallen victim to the creeping complexity that can be introduced when there is too much choice and too little control. Over time, different developers decided to use whatever “flavor of the month” tools looked interesting to them, and this resulted in a situation where it became next to impossible to debug existing use cases and create new ones.

Dell Technologies and Cardinality were able to quickly clean things up with the Cardinality ETL Engine, which provides an elegant and easy-to-maintain mechanism for ingesting data. The result is that the operator is now able to build use cases without having to worry about the complexity of ETL.

Data Indigestion

A variation on the complexity theme has to do with the complexity of data sources themselves.

Dell Technologies helped another customer that was saddled with having to keep up with a variety of data formats from different network probes. Having multiple probes is complicated by the fact that probe vendors occasionally change their data formats, requiring rework and telecom expertise to reformat the data into formats used for analytics. An additional problem is that some older, proprietary data formats can’t always be used with newer ingestion tools, introducing latency and performance limitations, and this ingestion “indigestion” can limit the kinds of real-time use cases that can be put into production.

By modernizing the customer’s environment with the Cardinality ETL Engine, we were able to relieve the customer of the headache of having to manage a multitude of data sources and were further able to vastly improve streaming performance. The number of data records ingested and parsed per day increased from 9 billion to 23 billion and the number of files needing to be discarded due to format quality issues dropped to nearly zero.

“Plumbing” Matters

The Dell Technologies Service Provider Analytics with Cardinality solution dramatically reduces customers’ data ingestion pain points with an ETL Engine that allows customers to:

◉ Get into production fast with “out-of-the-box” ETL functionality that is purpose-built for telecom environments

◉ Collect streaming and non-streaming data with the low latency and high throughput

◉ Lower OPEX by reducing the resources needed to manage multiple data formats from different sources

◉ Scale from small to huge on a unified data analytics platform

Dell Technologies and Cardinality offer customers a Kubernetes-based microservices platform that spans the data pipeline from ETL to analytics to prebuilt telecom use cases is tuned to run on scalable, high-performance Dell EMC PowerEdge clusters and is integrated with Dell EMC Isilon and Pivotal Greenplum. Together, Dell Technologies and Cardinality are committed to ensuring they can make the most out of data analytics.

Continue reading

In A Data-Driven World, Innovation Changes Are Forcing a New Paradigm

For the last two decades, I’ve enjoyed working at Dell Technologies focusing on customer big-picture ideas. Not just focusing on hardware changes, but on a holistic solution including hardware, software, and services that achieves a business objective by addressing customer goals, problems, and needs. I’ve also partnered with my clients on their transformation journey. The concepts of digital transformation and IT transformation have been universal themes and turning these ideas into realities is where the rubber meets the road.

Now as I engage with customers and partners about Microsoft solutions, an incremental awareness of the idea of “data”, and how data is accessed and leveraged, has become evident. A foundational shift around data has occurred.

We are now living in a new era of data management, but many of us were not aware this change was developing. This has crept up on us without the fanfare you might see from a new technology launch. When you take a step back and look at these shifts in their entirety you see these changes aren’t just isolated updates, but instead are amplifying their benefits within each other. This is a fundamental transformation in the industry, similar to when virtualization was first adopted 15 years ago.

For many, this change started to become apparent with the end of support for SQL Server 2008 earlier this year (along with support for all previous versions of the product). This deadline, coupled with the large install base that still exists on this platform, is helping the conversation along but it’s not just a replace the old with the new in a point-by-point swap out. The doors opened in this new era force a completely different view and approach. We no longer need to have a SQL, Oracle, SAP, or Hadoop conversation – instead it becomes a holistic “data” point of view.

In our hybrid/multi-cloud world, there is not just one answer for managing data. Regardless of the type of data or where it resides, all the diverse data languages and methods of control, the word “data” can encompass a great deal.

Emerging technologies including IoT, 5G, AI and ML are generating greater amounts and varied types of data. How we access that data and derive insight from it becomes critical, but we have been limited by people, processes, and technology.

People have become stuck in the rut of, “I want it to be this way because it has always been this way.” Therefore, replacing dated/expired architectures becomes a swap out story verses a re-examine story and new efficiencies are completely missed. Processes within the organization become rigid with that same mindset and, dare I say politics, where access to that data becomes path- limited. Technology is influenced by both people and process as “the old way is good enough, right?”

The value/importance of “data” really points back to the insight that you drive from it. Having a bunch of ones and zeros on a hard drive is nice but what you derive from that data is critically important. The conversations I have with customers are not so much, “Where is my data and how is it stored?” The conversation is more commonly, “I have a need to get business analytics from my proprietary data so I can impact my customers in a way I never did before.”

To put my Stephen Covey hat on, we are in a paradigm change. What is occurring is incredibly impactful for how customers should view and treat data. There are three key areas that we will examine with the new paradigm today and we’ll start with data gravity.

Data Gravity

Data gravity is the idea that data has weight. Wherever data is created, it tends to remain. Data stores are getting so big that moving data around is becoming expensive, time constrained, and database performance impacting. This in turn, results in silos of data by location and type. Versioning and lack of upgrade/migration/consolidation of databases also perpetuates these silo challenges.

As with physical gravity, we understand that data’s mass encourages applications and analytics to orbit that data store where it resides. Then, application dependency upon the data’s language version cements the silo requirement even further. We have witnessed the proliferation of intelligent core and edge devices, as well as bringing applications to that place where the data resides – at the customer location.

Silos of data based on language, version, and location can’t be readily accessed from a common interface. If I am a SQL user, how do I get that Oracle data I need? I cannot just pull all my data together into a huge common dataset – it’s just too big. We see these silos in almost every customer environment.

Data Virtualization

This is where data virtualization comes into the story. Please note this is not a virtual machine (a common confusion on the naming). Think instead of this being data democratization: the ability to allow all the people access to all the data – within reason, of course. Data virtualization allows you to access the data where the data is stored without a massive ETL event.  You can see and control the data regardless of language, version, or location. The data remains where it is, but you have real-time source access to this data. You can access data from remote or diverse sources and perform actions on that data from one common point of view.

Data virtualization allows access into the silos that, in the past, have been very rigid, blocking the ability to effectively use that data. From a non-SQL Server point of view, having unstructured data or structured data in a different format (like Oracle), required you to hire a specialized person with a specific skill set to access that data. With data virtualization, that is no longer a barrier as these silo walls are reduced. Data virtualization becomes data democratization, meaning that all people (with appropriate permissions) can access and do things with that data.

From a Microsoft point of view, that technology came into reality with Polybase. Polybase with SQL Server allows access with T-SQL, the most commonly used database language. I started using this resource with the Analytics Platform System (APS) many years ago. After Microsoft placed this tool into SQL Server in 2016 and updated its functionality tremendously in SQL Server 2019, we now can ingest Hadoop, Oracle, and use orchestrators like Spark, to access all these disparate data sources. To visualize this, think of Polybase with SQL Server 2019 as a wrapper around these diverse silos of data. You now can access all these disparate data sources within one common interface: T-SQL using Polybase.

Holistic Solution

The final tenet of this fundamental change is the advent of containerization. This enablement technology allows abstraction beyond virtualization and runs just about anywhere. Data becomes nimble and you can move it where needed.

It’s amazing how pervasive containers have become. It’s no longer a science experiment, but is quickly becoming the new normal. In the past, many customers had a forklift perception that when a new technology comes into play, it requires a lift and replace. I’ve heard, “What I am doing today is no longer good, so I have to replace it with whatever your new product is, and it will be painful.”

I’ve been using the phrase that containerization enables “all the things”. Containerization has been adopted by so many architectures that it’s easier to talk about where you can’t do it verses where you can. Traditional SAN, converged, hyperconverged, hybrid cloud — you can place this just about anywhere. There is not just one right path here — do what makes sense for you. It becomes a holistic solution.

There are multiple ways to address the business need that customers have even if it’s leveraging existing designs that they’ve been using for years. Dell Technologies has published details of several architectures supporting SQL Server and has just recently published the first of many papers on SQL Server in containers.

The answer is, you can do all these things with all these architectures. By the way, this isn’t specific to Microsoft and SQL Server. We see similar architectures being created in other databases and technology formats.

These three tenets are each self-supporting to the new paradigm. Data gravity is supported by data virtualization and containerization. Data virtualization allows silos when needed (gravity) and is enabled by containerization. Containerization gives access to silos (wrapper) and is the mechanism to activate data virtualization.

From a Dell Technologies point of view, we are aggressively embracing these tenets. Our enablement technologies to support this paradigm are called out in three discrete points – accelerate, protect, and reuse. We will review these points in a separate blog.

There is much more to come as we continue this journey into the new era of data management. Dell Technologies has deeply invested in resources around this topic with several recent publications and reference designs embracing this paradigm change. Our leadership on this topic is the result of our 30+ year relationship with Microsoft and our continuing “better together” story.

Continue reading

AI Scaling and Other Musings

Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) are more than just flare for an expo booth. Companies are transforming their business practices, driving productivity, and creating new opportunities with the power of data and AI. Faced with limited resources, organizations must decide whether to scale their AI processes up or out. This key decision will better leverage the talent of an organization’s data scientists, capturing operational efficiencies, transforming decision making, and delivering stronger business results.

So, what are AI, ML, and DL, and how are they related? DL is a subset of ML, and ML is subset of AI – clear as mud right? AI can independently sense, reason, act, and adapt. ML and DL use mathematical algorithms to ‘learn’ as opposed to being explicitly told what to do as in expert systems (imperative programming using a lot of ‘if, then else’). ML techniques use a variety of algorithms to create mathematical models which can then be used to predict an output based on some new input. DL structures algorithms in layers to create an artificial “neural network” which can learn and improve with vast amounts of data to make a better output (a decision) based on some new input.

Some characteristics of ML and DL are shown below in Figures 1 and 2:

Figure 1

Figure 2

As an example of ML and DL, think of the problem of predicting someone’s weight (Figure 3). It could be a simple data set with height and weight then, using ML, you come up with line of best fit. Conversely, you could have tons of image data labeled with weight, then DL could refine the model with many more considerations, such as facial features, therefore improving accuracy.

Figure 3

The fuel of AI is lots of data; advanced models, algorithms and software; and high-performance infrastructure. As a general rule, if you are fortunate and have all three fuels, your organization should use DL. If you have two out of three, one, or none, use ML.

The AI journey is a challenge. When we talk to our customers, they cite limited expertise, inadequate data management, and constrained budgets. Their business challenge is threefold: Define, Develop, and Deploy. They must define the use cases, requirements, business goals for artificial intelligence. Then, they must develop the data sources, infrastructure, trained models, and refinement process surrounding an AI system. Finally, they need to deploy the AI-enabled systems with tracking tools and inferencing, at scale.

Now that we have covered the seemingly daunting challenges of implementing AI, let’s dive into the upbeat topic of a scale up or scale out strategy for AI. Scaling up implies adding more resources to an existing system. In the case of AI, that means how many domain specific architecture elements (DSA) are in an individual server. DSA is becoming the code name to mean anything that accelerates beyond general purpose computing, such as GPUs, ASICs, FPGA, IPUs, etc.  Scaling out, however, means adding more systems, each with some number of GPUs. Think of this in server terms as a 32-socket server vs 2-socket server. This is like the 30+ year old argument of mainframe vs x86 scale out. Scale out economics emerged the clear victor.

For example, consider a family with independent transportation needs. The family could buy a Hennessy Venom GT – the fastest car on the planet – and one by one they could share it to rapidly meet their transportation needs. Or, they could buy n number of sufficient-performance cars and meet all their needs all the time.

Figure 4

Just as the family will simultaneously need to take concurrent trips, corporations on the AI journey will need more than one data scientist. Our job in technology is to make all data scientists productive all the time.

The first set of data below uses MLperf (MLperf builds fair and useful benchmarks for measuring training and inference performance of ML hardware, software, and services) across the Dell portfolio of domain specific architecture-enabled (DSA) servers. In Figure 5, you can see the scaling from 2 V100 GPUs in the Precision 5820 to 4 V100 in a PowerEdge C4140 to 8 V100 in the DSS 8440 to 8 in an NVLink-enabled server. The 8-NVLink server performs slightly better than the 8-PCIe-enabled DSS8440 (note the DSS8440 can support ten V100’s).

Figure 5

If you look closer at the individual 8-GPU scores below in following two figures, the scores are very close with exception of translation and recommendation for the two different 8-GPU cases. That said, the performance gap has closed between NVLink vs PCIe with proper PCIe topologies and better PCIe switch peer-to-peer support, and it will close more with PCI data speeds of Gen4 and Gen5 which is 16GT/s and 32GT/s respectively vs Gen3 which was 8GT/s.

Figure 6

Figure 7

If we take the same MLPerf, compare an 8-NVLink server to 2x 4-node NVLink servers, and spread the sub benchmarks across the two servers, we can now get all the work done faster than a single 8-NVLink server (Figure 8).

Figure 8

This checks out because, according to Amdahl’s law, the speed-up is not linear in large scale parallelism. Thus, it is important to pick a point on the speed-up scale before it gets into the non-linear region and to stay in the scale out economic sweet spot. Figure 9 shows scaling for 1-to-4 and 1-to-8 GPUs, as you can see, 1-to-4 GPU scaling has higher efficiency across the board (Figure 9).

Figure 9

Mark Twain popularized the saying, “There are lies, dammed lies, and statistics.” If he were still a journalist in the Bay Area today, he would turn our attention toward the pitfalls of benchmarks. Benchmarks in the right context are fine, but reduced to simple metrics, benchmarks can cause serious problems. Leaders must understand the underlying drivers of benchmarks to make informed decisions. Continuing our scale out or in example, 2x scale out nodes outperform a single scale up node using MLperf, the best industry benchmark. The scale out C4140 is also the best MLperf-per-dollar option for AI, which reflects the optimization point of using economic scale out nodes. Additionally, if you noticed in the news, VMware recently announced the acquisition of Bitfusion, which is a scale out approach to elastic AI infrastructure.

Now, this is not to say 8 GPU+ systems don’t have their place, just like we still have 8 to 32-socket x86 servers, they have their place and need. But there is an economic law and Amdahl law at play which suggest that the economics of scale out win except in the corner cases. Scaling GPUs is impacted by two main variables: data size and optimized code. If the data set is too small and scaling inefficiently with more GPUs, you end up shuffling parameters between GPUs more than doing computation on the data. Most models and use cases in practice for DL in enterprise will be best run on one GPU because of dataset sizes and quality of code. AI-proficient companies have hired skilled “ninja” programmers due to use cases/datasets that require training across 8+ GPUs to meet time-to-solution requirements. Everybody else, good luck with getting better performance on two or more GPUs than a single GPU.

The fact is this, companies on an AI journey will have tens or hundreds of data scientists, not just one. It is our job in technology to make sure all data scientists are as productive as possible. If one data scientist finishes her job fast while the others are drinking coffee and waiting for their turn, the company is not productive. Much like buying a super car to share with your spouse, only one person can be happy at a time. As monk and poet John Lydgate wisely wrote, “You can please some of the people all of the time, you can please all of the people some of the time, but you can’t please all of the people all of the time.” Well, in this case you can make all the data scientist happy all the time by enabling productivity to all, all the time – by investing in scale out and proper GPU workload orchestration.

Continue reading