The U.S. healthcare industry has more than 12 million diagnostic errors each year. Life expectancy is not improving, costs are too high, and physicians are overworked. And there is more medical data per patient than ever before. The good news, however, is that from medical imaging to analyzing genomes to discovering new drugs, artificial intelligence (AI) in healthcare is getting better and more sophisticated at doing what humans do, and doing it more accurately, more quickly, and with lower cost. A McKinsey review in 2018 predicted healthcare as one of the top five industries with more than 50 use cases that would involve AI, and over $1 billion has already been raised in startup equity.
Doctors are now making faster, more accurate diagnoses thanks to AI. Of heart patients, 61% are avoiding invasive angiograms, cutting treatment costs by 26%. AI is reducing diagnosis errors in breast cancer patients by 85% and enabling MRIs to accelerate image reconstruction by a factor of 100—with 5-times greater accuracy.
AI allows drug researchers to expedite discovery and development and can reduce the cost of bringing new drugs to market during their 12–14 years of development. AI analyzes millions of molecules to quickly identify potential drugs and lower development costs, and researchers of Alzheimer’s, cancer, and multiple sclerosis drugs report a tenfold increase in their productivity.
As a result of AI, healthcare costs are falling and outcomes are improving. The cost of AI-automated breast cancer risk assessments is 5% lower compared with current genomic tests. Two million stroke patient neurons are being saved each minute by rapid AI diagnoses, and tens of millions of healthcare professionals globally will use AI and retina imaging to quickly detect countless medical conditions.
Today’s data visionaries are joining NetApp and NVIDIA to apply AI and deep learning to healthcare’s greatest challenges. Together, we are accelerating medical discoveries and improving patient care.
With so many “next-generation” data management solutions available today, why would anyone choose SAN (Storage Area Network) over something like HCI or just moving everything straight to the cloud?
I see SAN as a solution to a problem. SAN takes away the complexity of a multi-server environment by consolidating storage onto a single, high-performance appliance that is redundant and accessible over a private network.
This architecture is also useful if you have already invested significantly in compute but need to scale storage on demand. With SAN, you can fully maximise compute capacity while scaling storage capacity independently, squeezing every compute cycle and GB out of your investments.
Pooling resources together offers some cost benefits as well—there is economy in scale after all! Take, for example, an implementation with 20 independent VMware data stores, each with around 150GB of free space. While there is plenty of compute, the VM requires ~200GB of storage capacity. In this instance, you would have to purchase a new server to accommodate just a single VM, even though there is around 3TB worth of unused storage space just sitting there.
What if the CPU utilization is around 30% on all servers and RAM is at 40%, but there are no more HDD slots? You could maybe put in new disks to compensate, but what do you do with the VMs while you perform this upgrade? SAN enables you to seamlessly migrate workloads from one server to another and perform maintenance of underlying systems without any disruption. With SAN, workloads of different sizes can even sit on the same shared storage but utilize different compute.
By choosing to put NetApp at the heart of your SAN infrastructure, you get many other great features:
Storage efficiencies such as data deduplication, compression, and thin provisioning help free up capacity and relieve management strain
Data protection features like snapshots, SnapMirror, and SnapVault give you the ability to offload computation cycles for analysis and data protection to another system while keeping your primary storage free, in addition to helping you stay compliant
Data replication technologies such as MetroCluster allow you to mirror two NetApp systems over a distance of 700 kilometres for geographically disparate disaster recovery
I wanted to touch on another big benefit of SAN: speed! With the advancements in flash technology and the emerging popularity of NVMe (non-volatile memory express) SAN is able to offer you higher IOPS and lower latency access for your business-critical applications.
This speed of access is especially important given the ever-increasing demand for instant access to the data generated by our business-critical applications. Often these applications are designed for a SAN environment, and as such, perform better in a SAN environment over other solutions like cloud.
SAN is not just for those continuity systems, but also for the modern data centre, providing us with ultra-low-latency access, storage efficiencies, and data retention across the data fabric, ensuring our data is where we need it, when we need it.
The artificial intelligence (AI) survey data discussed in the blog post is based on answers from more than 100 customers at recent AI conferences to gather primary research on AI data management challenges, AI infrastracture and tools, and reasons for choosing different types of data storage.
The majority of survey respondents were software developers (24%), data scientists (18%), researchers (15%), line-of-business owners (7%), data analysts (6%), or general IT (6%). Company size of respondents ranged widely: more than 10,000 employees (45%), fewer than 1,000 employees (28%), 1,000 to 5,000 employees (18%), and 5,000 to 10,000 employees (9%). The top use cases for AI were healthcare and computer-assisted diagnosis, automotive and autonomous vehicles, manufacturing and robotics, and financial services and fraud detection.
The top three storage and data management challenges with AI were scaling storage, cloud integration, and backing up data.
The top three requirements when choosing an AI infrastructure vendor were cost, easy to deploy and manage, and services and support offerings.
The three most common tools used with AI were NoSQL databases, Apache Hadoop, and Splunk. The majority of respondents either use a Hadoop cluster (data lake) with AI or plan to in the near future.
The most common file systems used with AI are HDFS and NFS. Less than 10% of the respondents used ZFS and GPFS.
Cloud storage is the most popular storage used for AI, followed by direct-attached storage (DAS) and then external storage.
The top reasons for using servers with internal storage or servers with JBOD are performance, cost, and management decision. As shown in the following graph, the top reasons for using external data storage are performance, reliability, and data protection.
NFS is the protocol of choice for AI with external storage, followed by Fibre Channel and then NFS. Ease of use, easy to scale, cost, and already use cloud for compute are the main reasons for using cloud storage. The most popular services used in the public cloud are Amazon Web Services, Microsoft Azure HDInsight, and Google Cloud Dataproc.
Data center technology moves in cycles. In the current cycle, standard compute servers have largely replaced specialized infrastructure. This holds true in both the enterprise and the public cloud.
Although this standardization has had tremendous benefits, enabling infrastructure and applications to be deployed more quickly and efficiently, the latest computing challenges threaten the status quo. There are clear signs that a new technology cycle is beginning. New computing and data management technology are needed to address a variety of workloads that the “canonical architecture” has difficulty with.
NetApp and NVIDIA share a complementary vision for modernizing both the data center and the cloud. We’re using GPU and data acceleration technologies to address emerging computing workloads like AI, along with many other compute-intensive and HPC workloads, including genomics, ray tracing, analytics, databases, and seismic processing and interpretation. Software libraries and other tools offer support to teams moving applications from CPUs to GPUs; RAPIDS is one recent example that applies to data science.
Server Sprawl and the Emergence of GPU Computing
Server sprawl is a painful reality in many data centers. Even though CPUs get more powerful every year, the total number of servers keeps climbing because:
More CPUs are needed to support the growth of existing workloads
More CPUs are needed to run new workloads
Digital transformation is accelerating the rate at which new application workloads are coming online, making the problem even worse. This is where GPU computing comes in. You’re probably aware that GPUs are being used for deep learning and other AI processing—as well as for bitcoin mining—but these are just the best-known applications in the field of GPU computing.
Beginning in the early 2000s, computer scientists realized that the capabilities that make GPUs well suited for graphics processing could be applied to a wide variety of parallel computing problems. For example, NetApp began partnering with Cisco, NVIDIA, and several energy industry partners to build GPU computing architectures for seismic processing and visualization in 2012. Today’s fastest supercomputers are built with GPUs, and GPUs play an important role in high-performance computing (HPC), analytics, and other data-intensive disciplines.
Because a single GPU can take the place of hundreds of CPUs for these applications, GPUs hold the key to delivering critical results more quickly while reducing server sprawl and cost. For example, a single NVIDIA DGX-2 system takes just 10U of rack space, cutting the infrastructure footprint by 60 times at one-eighth of the cost, compared to a 300-node CPU-only cluster to do the same work.
Data Sprawl Requires a Better Approach to Data Management
The same architectural approach that contributes to server sprawl also creates a second—and more insidious—problem: data sprawl. With the sheer amount of data that most enterprises are dealing with—including relatively new data sources such as industrial IoT—data has to be managed very efficiently, and you have to be extremely judicious with data copies. However, you may already have multiple, separate server clusters to address various needs such as real-time analytics, batch processing, QA, AI, and other functions. A cluster typically contains three copies of data for redundancy and performance, and each separate cluster may have copies of exactly the same datasets. The result is vast data sprawl—with much of your storage consumed to store identical copies of the same data. It’s nearly impossible to manage all that data or to keep copies in sync.
Complicating the situation further, the I/O needs of the various clusters shown in the figure are different. How can you reduce data sprawl and deliver the right level of I/O at the right cost for each use case? A more comprehensive approach to data is clearly needed.
Is the Cloud Adding to Your Server and Data Sprawl Challenges?
Most enterprises have adopted a hybrid cloud approach, with some workloads in the cloud and some on the premises. For example, for the workloads shown in the figure, you might want to run your real-time and machine-learning clusters on your premises, with QA and batch processing in the cloud. Even though the cloud lets you flexibly adjust the number of server instances you use in response to changing needs, the total number of instances at any given time is still large and hard to manage. In terms of data sprawl, the cloud could actually make the problem worse. Challenges include:
Moving and synching data between on-premises data centers and the cloud
Delivering necessary I/O performance in the cloud
You may view inexpensive cloud storage such as AWS S3 buckets as an ideal storage tier for cold data, but in practice it too requires a level of efficient data movement and management that may be difficult to achieve.
Tackling Sprawl with NetApp and NVIDIA
If you’re struggling with server and data sprawl challenges, the latest data management solutions from NetApp and GPU computing solutions from NVIDIA may be the answer, helping you build an effective bridge between existing CPU-based solutions and GPU-based ones.
NVIDIA leads the industry in GPU computing. GPU acceleration delivers results fast and reduces server sprawl. NVIDIA software tools make it easier than ever to get started.
NetApp helps you manage data more efficiently, eliminating the need for unnecessary copies. Data from dispersed sources becomes part of a single data management environment that makes data movement seamless. Advanced data efficiency technologies reduce your storage footprint and further reduce data sprawl. Data tiering allows you to deliver the right I/O performance for every workload, ensuring that GPUs aren’t stalled waiting for data.
NetApp® Data Fabric and NVIDIA GPU Cloud enable seamless and efficient use of the hybrid cloud. Together, the two companies enable a unified software stack from edge to core to cloud. In my next few blogs, I’ll examine the new technologies that will deliver the results you need, whether your workload is AI, analytics, genomics, or something else, while tackling the server sprawl and data sprawl challenges that threaten your operations—and your sanity. Upcoming topics include:
Unifying Machine Learning and Deep Learning Ecosystems with Data
The Promise of GPU Computing and a Unified Data Platform
Organizations around the world are growing their video surveillance solutions at a faster rate than ever before. With more cameras and higher bandwidth requirements comes the need for more storage, which can get costly very quickly. On the surface, traditional NVR-based or DVR-based video storage solutions may seem cost-effective when compared to enterprise-grade storage, but the more cameras you add, the more costly they become to manage. And it’s not just about the costs to own and manage that equipment. Do you know the true cost of downtime in your video surveillance environment?
What’s Really at Stake?
Prisons and correctional facilities around the world rely on video surveillance to keep their staff safe and guard against frivolous lawsuits. Using video, prison management can easily dismiss claims from inmates by proving that events never took place or happened differently than inmates describe. And by keeping tabs on inmate activities, security staff can make sure that their personnel are safe and monitored at all times. If their video surveillance goes down for any amount of time, the prison risks losing valuable evidence that can be used to protect the prison from legal penalties. Worse, failed surveillance can put their staff in harm’s way.
Manufacturing companies also rely on video surveillance to streamline operations and make sure that both staff and machinery are operating at maximum efficiency. This type of operational monitoring not only helps cut costs by reducing wasted time and effort, but it also means that the organization doesn’t need to hire as many managers to supervise factory floors and warehouses. Without available and resilient video surveillance, these companies lose out on valuable insight that they could use to run their businesses better and more cost-effectively.
Buy the Best and Only Cry Once
Off-the-shelf video surveillance storage solutions, although providing a low entry point, can often end up costing much more than enterprise-grade equipment over the lifecycle of the solution. With a traditional video surveillance storage solution, you need to add more and more DVRs to keep up with capacity growth. Cheap, white-box storage has a high rate of failure, meaning that equipment will need to be replaced more often, particularly when it is pushed to its limit in terms of capacity and performance. When those systems do need to be swapped out, you’re using up valuable IT hours that could be spent providing value to the business (to say nothing of the downtime incurred while the systems are undergoing repair). These costs add up, and, suddenly, that “cheap” solution isn’t looking so cheap after all.
With a proven video surveillance storage solution built on NetApp® E-Series storage systems, you can be sure that you’ll have the capacity you need, when you need it, without having to worry about losing critical data or wasting time maintaining your infrastructure. E-Series is a tested platform that’s purpose-built for your video surveillance solution. You get petabytes of capacity with always-on availability, so you never need to stress about adding more cameras or increasing your resolution
The lines between security and IT organizations are being blurred. With the rapid proliferation of video surveillance data and the rise of intelligent video management systems, security personnel now find themselves responsible for fast-growing IT infrastructure. In addition to keeping people and property safe, they now must also protect and manage terabytes—or even petabytes—of critical data.
Remember When Video Was Simple?
It wasn’t long ago that video was recorded directly onto the camera or fed into a central hub with closed-circuit television (CCTV). Yes, back then, video surveillance was simple—it wasn’t always reliable, but it sure was simple. As concerns around reliability, privacy, and centralization of data have become more prevalent, video surveillance has had to evolve. For many security operations, this evolution has resulted in an increasingly complicated infrastructure with myriad technical processes that require more time and effort to set up and maintain.
Who’s in Charge Here?
Ultimately, it is the responsibility of the security team to decide what video surveillance solutions to deploy and how to deploy them. But when it comes to the infrastructure to support those solutions, it is not always clear where that technology should live and who should manage it.
Enterprise storage can be foreign territory for the security organization. Security personnel would rather spend less time managing IT infrastructure and more time doing their jobs. However, at many organizations, IT doesn’t have the bandwidth to take on another complicated workflow. Furthermore, any additional budget is better spent upgrading cameras and adding capabilities—not adding headcount.
Making Enterprise Storage Simple Again
Five network video recorders (NVRs) are easy to manage. Fifty are a headache. When you get to 100 or more NVRs, it can be a nightmare. That’s the sweet spot for NetApp video surveillance storage solutions. NetApp® E-Series’ high-density, 60-drive shelves can support up to 558 cameras each. Instead of managing hundreds of devices independently, you can manage more than 270PB of storage from a single pane of glass.
NetApp E-Series systems offer purpose-built, high-performance, high-density storage, making them ideal for video surveillance workloads. E-Series storage is reliable, easy to deploy, and easy to manage, so you can grow your video surveillance deployment without adding complexity.
MAX Data does this how? MAX Data is software that runs on the application server and provides a file system that spans the PMEM and the storage tier. Applications stored on this filesystem get instant access to the data for both reads and writes. With MAX Data, you get vastly improved application performance with high throughput and ultra-low latency.
A new evolution of flash storage and a new class of memory is promising to change how applications access data, and NetApp MAX Data will enable and accelerate adoption of this exciting new technology.
This is critical because real time data requirements are increasing the load on applications and associated infrastructure. Our customers want to get the most out of their data; Response times and low latency are key attributes for an agile enterprise and the volume of data collected by organizations is growing exponentially.
Accelerate Your Business
With this huge explosion of data, companies are struggling to meet expectations across all application tiers due to slow, unpredictable data access impacting their customers, their IT staff, and their business. Upgrading to flash solve many performance challenges, but there are still customers looking to save tens to hundreds of microseconds especially when the gain is in the hundreds of thousands of dollars per microsecond (like in trading or retail solutions). Other vendors offer flash performance, but in silos, often without integrated data management, protection, efficiency and availability. While some improvements are noted, IT staff still wants and needs to bring their applications to the next level of performance and quality of service while reducing costs.
Oracle
Our early results show that you can get 60% more IOPS with four times less latency for your Oracle workloads. Another way to look at this would be to consider getting the same throughput using 40% fewer servers while still seeing four times lower latency. Your applications will also gain data persistency and enterprise-grade data services, making this a completely viable solution to enterprise applications.
MongoDB
The graphic below shows the power of MAX Data running in a MongoDB environment. This test shows results using a workload generator to query 7,000 documents 10,000 times in a MongoDB database. Each time, the query was randomly reading from a different range of 7,000 documents. This example can be viewed as mimicking a website like Reddit, where one subject can have thousands of threads and threads can have hundreds of comments. The result: MongoDB queries are completed three to ELEVEN times FASTER!
