Annual Update on Persistent Memory CXL: A Basic Tutorial
Guest Post

Cloud Workloads Driving Flash Adoption

Learn which workloads are impacting flash adoption and why, current trends, workload-related attributes to know about, what IDC means by cloud infrastructure and more.

00:20 Kuba Stolarski: Hi, thank you for joining this session on "How Cloud Workloads are Driving Flash Adoption." My name is Kuba Stolarski, I'm an analyst with IDC, a global market research company. I research topics such as compute and storage cloud infrastructure and workloads, and today I'm going to walk you through some ideas about which workloads are impacting flash adoption and why.

00:43 KS: But before we start, I'd like to talk for a moment about a hole in the ground. When you look at this photo, you might think I'm going to quote something like, "If you find yourself in a hole, the first thing you should do is stop digging." Or you might see the caption and wonder whether the shovel is the right solution for digging this particular hole. But, actually, I'm going to talk about a hole in the ground as a storage solution of sorts.

Imagine I have something valuable, maybe a giant bag of cash, and I want to store it some place where no one is likely to find it. I could employ this method and it might actually work as long as I can remember where I buried my treasure and no one saw me doing it. If I'm lucky, no one will find it by accident either. While this is technically a solution into my problem, there are other more effective, more efficient and even more lucrative ways to store my money. And that's what I'd like you to keep in mind. Ask yourself for the problems that you're trying to solve, for the applications that you need to deploy, what is the right solution for the job? The answer will not always be the same for everyone and for every solution.

01:49 KS: And in this presentation, we're going to look at what are some workload trends, and we'll look at a few workload-related attributes that you might want to consider as you try to answer this question for yourself, probably many times over. We'll look at what we mean by cloud infrastructure. What are workloads, at least the way IDC looks at them, and what is a way to look at workload requirements that relate to whether or not flash technology could provide the right solution for you? We'll see which workloads are growing, briefly touch on what we're seeing changing due to COVID-19 and, finally, a few comments about how we see workloads changing now and into the future. So, let's get started.

02:32 KS: First, let's take a look at how we view cloud infrastructure at IDC and how it relates to workloads, which we'll get to in a moment. We see this as a multidimensional thing. Workloads can be deployed on infrastructure owned by service providers or enterprises, and we've circled and color-coded where the service provider infrastructure is, usually referred to as off premises. We look at a variety of service providers, including cloud, digital service providers like Netflix and Twitter, hosters and managed service providers, as well as telcos. We even have a special category for the big eight, which we call hyperscalers. Next, workloads can be deployed in a shared or dedicated infrastructure environment, so that includes mostly on premises, but also includes dedicated hosting offerings like managed cloud.

03:29 KS: And, finally, we look at workloads and whether they are deployed on a cloud platform or not. And what is a cloud platform? Well, we defined it as a platform that has some specific attributes. It comes in the form of a shared standard service built for multi-tenancy, it's solution packaged, pre-integrated with required resources, it's self-service, it has elastic resource scaling and use-based pricing or metering, and it has a published API. And as you'll note on the bottom, cloud makes up about half of the infrastructure that was deployed last year, and this year it's on track to be more than half, in spite of, or perhaps even accelerated by COVID-19.

04:10 KS: Now, let's switch gears and talk briefly about the workloads themselves. What are they? They are a mechanism we use to segment servers and storage based on what is running on them. We take seven categories and break them into 18 workloads. Most of these workloads will map to types of software applications for which we have hundreds of categories and subcategories at IDC. The largest workload categories are business applications like CRM and ERM; data management, which includes structured and unstructured data; and IT infrastructure, which is like the backbone supporting the rest, the unsung heroes of the workloads world, like security and systems management. I'd also like to point out content and collaboration, including the tools many of us use to work together even at a distance, which has become critical this year. This is a category that has been growing considerably over the years.

05:08 KS: And here's a better relative view of how much infrastructure gets deployed in support of each of these workload categories. So, when we intersect cloud deployments and workloads, there are some commonalities that emerge. The workloads themselves are often distributed modular by design, running as stateless applications. The operating environments are almost always abstracted in some way, either through hypervisors or containers or both. They include automation and enable applications to run as microservices if they are architected that way.

And, finally, we see a commonality in business factors such as sensitivity of data, whether it's customer data, financial data, whether it's sensitive due to regulatory compliance requirements or due to competitive advantage. The necessary speed to deployment of new applications and application updates is often critical. And the scale or the ability to scale massively and across the globe is also becoming increasingly important.

06:10 KS: So, let's go back to the workloads and think about requirements. What do these workloads need to run? To run well, to run efficiently. In the context of storage media, we can encapsulate workloads as driving some combination of capacity, performance and cost requirements. But let's take out cost for a moment and focus on capacity versus performance as a dimension.

06:33 KS: We can think of capacity requirements in terms like total volume, which seems to always be growing tremendously, but also as density, if we think of data center space or footprint and efficiency. On the performance side, we can very explicitly narrow down the performance metrics to latency or how quickly you can start a request to I/O -- input/output -- or how many requests you can carry out in a period of time, often measured in IOPS -- input/output operations per second -- and throughput or how much data you can send in or out, or how fast. And, of course, if you didn't know before, you'll know after this event that flash technology is very well-suited for meeting the performance requirements, even as its ability to meet capacity-related requirements is also improving over time.

07:27 KS: Now, workloads are not going to map to these requirements one-to-one. You could have two instances of the same database platform, for example, with very different requirements that are unique to what you're trying to do with that database. Is it transactional? How many transactions do you need to process? If it's analytical, do you need real time insights or are periodic reports fine? While mapping one to-one is impossible, there are trends and here they are. At the extremes of the capacity-based requirements, we see workloads like file and print, some business applications and web serving. At the other end on the performance-focused side, we see unstructured data analytics which is increasingly tied in with AI/ML use cases and VDI -- or virtual desktop infrastructure -- remote desktops.

08:18 KS: And one thing you'll hopefully see is that the growth that we expect in the workloads closer to the right tends to be greater than the growth we expect to see in the workloads on the left. And that's a significant statement, as the reality of workloads is that they don't really spike relative to each other as much as you might think from listening to the hype. Just about everyone runs many of these workloads and sooner or later they need to be refreshed, so they rarely ever really decline.

The differences you see here, from about 50% to 70% cumulative growth on the right, compared to 20% to 50% growth on the left, is actually like night and day. But this is a static view from a requirements perspective. Requirements are changing all the time as well, and they are not balancing against each other. Instead, capacity and performance requirements for workloads are constantly increasing. However, one of the reasons we are seeing flash adoption and why it will continue, is that price capacity for flash is improving and significantly, which is making it much easier as time goes on to use flash technologies in an increasing number of workload instances. You could interpret that to say that flash is winning the long game.

09:38 KS: And, finally, there is a workload of sorts that isn't really a workload, but it's all of them. Workload consolidation -- which today usually takes the form of a cloud platform deployment and is increasingly manifesting as hybrid and multi-cloud deployments -- is driving more demand for flash, even when many of the workloads being consolidated onto one cloud platform are capacity-oriented. The reason is that together these workloads can wreak havoc on storage, I/O and throughput, so flash is increasingly used to alleviate those aggregated performance bottlenecks in the consolidated environment.

10:14 KS: We see this as an important game changer for how workloads and workload management are influencing infrastructure choices already today, but even more so in the future. So, we're near the end here, and I have a few points to make about what we are seeing with COVID-19.

It's obviously early to tell what the long-term impact will be. However, we know that overall the impact has been generally negative for enterprise and infrastructure deployments this year. On the other hand, cloud, especially public cloud workloads, have actually thrived in this environment. In particular, we see collaboration media and VDI or remote desktop reaching new levels of usage in the cloud this year. How will those impact flash? Well, we'll see, but these are performance-intensive workloads -- even the ones within collaboration that are being utilized during the pandemic -- so one could easily infer that flash is again playing an important role in how infrastructure is being deployed for these expanding workloads.

11:15 KS: Finally, let me just touch on a few things. Workloads are useful, but as we saw, they don't tell us the full story. There are macro trends in the industry that cross multiple workloads that will contribute to the need for more flash. Applications are increasingly becoming real time, cloud-native or developed with new cloud capabilities in mind, distributed geographically -- we call that edge computing -- and heterogeneous from a resource perspective. What that means is that internal functions can now be separated out and different resources, whether processors or media, can be used to optimize overall performance of the application. We're also seeing more composite or complex workloads that cross boundaries across multiple, traditional core workloads. Some examples of these are shown here, such as AI/ML, media applications and services, virtual network functions and massively parallel computing.

12:17 KS: So, in summary, we've gone through some ideas about how to think about the intersection of workloads, cloud and flash technology. Ask yourself whether flash, as well as other tools, are the right tools for the problems that you are trying to solve. When deploying cloud workloads, consider their performance-centric requirements and how they are tied to the architecture, operation and business needs such as data sensitivity, speed and scale. Remember that many of the cloud workloads that you might be deploying or expanding over the next few years are also many of the same workloads that benefit most from low latency, high IOPS and high throughput.

And, finally, keep in mind that workloads are just a piece of this. Macrotrends and complex composite workloads are what's coming, and they will push performance to its limits. This will open the door to even greater flash technology adoption in its many forms.

13:14 KS: With that, I thank you for your time. Please stay safe. And enjoy the rest of the event.

Dig Deeper on Flash memory and storage

Disaster Recovery
Data Backup
Data Center