Compute and Cloud Object Storage: A Unique Market Evolution

Conversations and considerations around cloud Infrastructure-as-a-Service (IaaS) often begin with compute solutions. How many cores or VMs does my workload require? Are they memory intensive? GPU intensive? Do we need general purpose machines, bare metal, both? These are some of the first questions asked when it comes to cloud infrastructure support for critical applications and workloads. However, this conversation is far from complete without the second half of the infrastructure equation: storage. The resource that helps organize, secure, serve, and manage growing volumes of application data.

Similar to computing resources, storage also comes in a variety of shapes and sizes. And as the cloud IaaS market has grown and matured, we’ve seen major providers standardize around a core set of compute and storage offerings. Each of these services and their key differentiators are described briefly below:

Cloud Compute Service Types:

Cloud computing services are broadly categorized into three groups based on the type of compute technology they are built on (e.g., CPU vs GPU), and various forms of deployment (e.g., multitenant vs dedicated):

1. Virtualized Compute Instances: your classic VMs packaged into a services delivery model. Typically comprised of x86-based CPUs (but don’t forget ARM), users choose the number of cores/threads they need and the amount of memory and attached persistent storage needed for their applications. In many cases these instances are pre-configured and advertised as “general purpose,” “compute optimized,” or “memory optimized” based on the need. Virtualized compute is the workhorse of the industry and serves the broadest range of application workloads.
2. Accelerated Compute Instances: like virtualized compute, accelerated compute services use GPUs and/or APUs as the primary form of computing power, instead of x86-based processors. These instances will also be pre-configured for different memory and persistent storage options. The main differentiator is the level of performance accelerated compute provides for specific workloads. You will typically see accelerated compute instances leveraged in machine learning, artificial intelligence, and a range of analytics workloads.
3. Bare Metal Instances: offer a range of non-virtualized computing, memory, and integrated storage. This often gives the user more flexibility over their compute and memory configurations, but more importantly provides a more dedicated cloud computing environment. This is because bare metal instances are often configured on a more bespoke, single-tenant basis, which results in lower (or no) utilization of pooled, multitenant resources. The major tradeoff here is cost. Bare metal instances are significantly more expensive to rent compared to traditional virtualized instances.

Cloud Storage Service Types:

Cloud storage services are primarily defined by protocol. And although some storage services or different tiers of service will use a mix of media types, this detail is usually several layers of abstraction away from the buyer. The result is that most organizations choose a storage solution based on protocol, price, and various performance SLAs, rather than media type.

1. Object Storage: services that utilize object-based storage architecture (characterized by the creation of objects, object metadata, and storage buckets) to store primarily unstructured data which is accessed via web API protocols.
2. File Storage: synonymous with Network Attached Storage (NAS) in the on-premises world, cloud file storage services that provide hierarchical access to stored data for users and applications (characterized by creation of files and folder structures), primarily using file sharing protocols such as NFS and SMB.
3. Block Storage: synonymous with Storage Area Network (SAN) in the on-premises world, cloud block storage services store data volumes on blocks, which are typically deployed as the core persistent storage layer attached/mounted directly to a compute instance via SCSI, fibre channel, or SATA.

The Growing Applicability of Cloud Object Storage

Historically, object storage was confined to secondary storage workloads (e.g., backup, archive) which land in the “cool” category of the enterprise cloud storage landscape. This isn’t necessarily a bad thing. In fact, a large proportion of enterprise stored data falls into this “cool” category, as illustrated below:

The Enterprise Cloud Storage Landscape: From Cold to Hot

Today, most organizations storing data in the cloud expect it to be available at the click of a mouse. This is one of the main reasons driving our assumption that the majority of stored data in the cloud (approx. 60% by volume) sits within the cool/warm/hot category; characterized by accessibility, low latency, and wide distribution of iterative data. This is where cloud object services like Wasabi thrive, and while it is important to acknowledge use cases in this category exists on a spectrum, we’re committed to offering a single tier of “hot” storage for everything.  The principal requirement of hot object storage is that data should be available at an instant, meaning 1/10^th of a second or less. Beyond millisecond access to your data, what counts most is to customers price, ingest and egress speeds, the physical location of the storage (for both speed and data sovereignty reasons), and durability (the statistical probability of losing data).

As we think about the evolution of the cloud storage market, I argue it is this 60% of enterprise data placed in the “cool/warm/hot” segment which has quickly become a significant opportunity and battleground for cloud object services. This segment is often comprised of data and storage for various business applications (e.g., email, file sharing apps, enterprise resource planning, CRM, supply chain management, human capital management, etc.), as well as a growing range of content and content delivery applications (e.g., media streaming, web serving, video and image sharing) which digital enterprises rely on.

As the cloud storage market continues to grow and mature, we expect object services to increasingly gain share by doing what object storage does best: delivering the ideal mix of price, performance, and security compared to alternatives. However, there are challenges. As object storage is utilized for a growing range of “hot” workloads, it often comes with a requirement to directly connect compute services and resources – whether that be virtual machines, accelerated compute, or bare metal instance configurations. This ongoing marriage of “hot” object storage capacity and various forms of compute resources will be a critical component of IaaS market evolution. And we won’t have to wait long to experience the effects. The rapid ascendance of artificial intelligence and machine learning solutions in commercial and business environments has the potential to upend our expectations regarding the need to leverage compute resources alongside massively scalable volumes of performant and cost-effective unstructured data. It also has the potential to expand the use cases and volumes of data within the “hot” segment of the stored data spectrum, resulting in an even higher proportional mix of enterprise stored data within the segment.

Acknowledging the Elephant in the Data Center: Generative AI

Source: Image created using OpenAI DALL·E 2

Blockchain and the Metaverse are old news thanks to the commercial and newsworthy successes of ChatGPT at the end of 2022. Halfway through 2023, the cat is completely out of the proverbial bag, and solutions based on generative artificial intelligence (AI) have been tagged as the next big thing in technology from both enterprise and commercial standpoints.

However, it remains very early days for generative AI as a technology, especially when it comes to its applicability within the enterprise. Despite short-term uncertainty, demand for generative AI-based solutions and services are already having an impact on downstream hardware markets. Just look at NVDA and AMD – two providers being primed to capitalize on the insatiable growth for generative AI solutions by supplying the accelerated compute hardware required to power a new generation of solutions and services. The floodgates are also opening at the software level, with generative AI models becoming available as commercial API services for application developers. This API accessibility will drive test and development of new solutions and speed time to market for enterprises looking to be on the bleeding edge of generative AI technologies.

Ultimately, what we’re describing here is the explosive growth of a new technology segment that most certainly will provide a boon to the cloud infrastructure services market. Much of the application development, hosting, and deployment of generative AI solutions and services will be done on/in the public cloud. For many enterprises, deployment of public cloud compute and storage services will be the quickest and most cost-effective way to procure the infrastructure resources needed to develop solutions, quickly get them to market, and generate a competitive advantage in this nascent segment.

It is too early to attempt to quantifiably measure the impact of generative AI solutions on the IaaS market. Certainly, it will drive growth, but how much growth should we attribute to this new technology? We believe that object storage, for many of the reasons illustrated above, is positioned to best address the range of data scalability, accessibility, and ease-of-use requirements for this emerging technology. However, it’s important to remember we’re dealing with a nascent category, and prudence might be our best virtue as we slowly uncover real-world use cases and weigh them against the current rate of hype surrounding all things generative AI.

From a big-picture market perspective, I think we can say with confidence that adoption and growth of generative AI solutions will change the way we think about the relationship between various types of cloud compute and storage resources; emerging generative AI-based use cases will demand mass quantities of traditional and accelerated compute, connected to equally high performant and massively scalable volumes of storage. As generative AI matures as a technology, and as enterprises discover, develop, launch, and improve services, they will also work to operationalize this new class of AI/ML/analytics-based solutions at scale and make them as cost-effective as possible. And this is where modern cloud storage services – specifically object storage – will have the opportunity to shine, by eliminating cost and fee complexity, meeting performance and scale requirements, and providing a sustainable storage foundation to build on.

Do your mission-critical cloud applications and workloads require computing power? Wasabi’s S3-compatible object storage can be quickly and seamlessly integrated with compute services of your choosing. Furthermore, Wasabi partners with leading cloud compute providers like Equinix Metal, Hivelocity, IBM, Vultr, Sushi Cloud and others, to ensure you can connect Wasabi storage volumes to virtualized compute, accelerated compute, bare metal configurations, and anything else your workloads require.