The Technical and Cultural Implications of ‘as a Service’ anywhere
Anywhere [1] computing, the promise of portability in Information Technology, has been a buzzword for decades. Why did this concept grab our attention and continue to do so? Where did it start and why is it still relevant today? In this post, I will discuss the origins and history of this term, its importance, and why we should still care about it going into the 2020s. It is my belief that the 2020s will see two major movements with anywhere computing:
- Hybrid multi cloud: heterogeneous interoperable public and private clouds. [2]
- ‘As a Service’: acceleration of software delivered as standardized “as a service”.
First, I will go into the origins of anywhere computing, and then I will discuss the cloud generally and the tough decisions customers must make when deciding on their cloud strategy. Additionally, I will discuss where popular vendors started their “as a service anywhere” journey and what strengths and weaknesses they have. Finally, I will explain my hypothesis as to why I believe “as a service anywhere” is everyone’s future.
History
The earliest incarnation of the anywhere concept began with the move away from the centralized homogeneous Glasshouse Computing of the 1960s to the rise of UNIX® [3] workstations in the 1970s. From then on, the move to increasingly portable computing continued with the rise of PCs in the 1980s, which accelerated heterogeneity in computing. The Java programming language, created in 1991, offered the promise of the true anywhere benefit of ‘write once, run anywhere’.
The first decade of the 21st century brought about the explosive growth of the Internet. Suddenly, anyone, anywhere, could communicate with billions of others, anywhere over TCP/IP. In the last decade, we have seen both the trend towards the Glasshouse Computing paradigm with the rise of homogeneous and often proprietary public clouds and the trend towards a true computing everywhere with Internet-connected edge devices and smartphones that made computing almost ubiquitous.
While in the 1980s, PCs promised economic value from “computing at your fingertips”, promising independence from the slow and bureaucratic ivory towers of corporate IT, subsequent modernizations brought more extensive changes to enterprise IT:
● Hypervisors significantly reduced capital expenses
● Application-server frameworks such as Java 2 Enterprise Edition® [4] increased developer and operator effectiveness
● Micro services and agile methodology increased product owner and developer velocity
● Containers such as Docker increased velocity of developers delivering value to production and using Kubernetes as the platform, or middleware, enabled management of containerized software at scale at relatively low cost
The technological, organizational, and cultural trends above changed how software was created and delivered to production. New learnings and technological advancements further reduced capital and operational expenses and increased developer effectiveness, and these processes continue in the present time. Consequently, these innovations can directly impact how much economic value a creator and consumer of software can achieve.
The Cloud
“According to Gartner, through 2024, nearly all legacy applications migrated to public cloud infrastructure as a service will require optimization to become more cost-effective.” [5]
Application optimization is most often done in stages. The first stage is lift and shift to public cloud infrastructure as a Service. The second stage is replacing storage services like databases with databases as a Service. The third stage is a refactoring of the application to run on elastic platforms such as Kubernetes as a Service. “As a service” is the optimization medium used to benefit from seemingly infinite capacity and pay per use.
Cloud computing revolutionized the enterprise IT environment. No longer did server rooms have to be maintained on premise by a team of dedicated devOps specialists. In the last decade, businesses have primarily moved to using public clouds. These public clouds have generally optimized their platforms for two primary use cases:
1. ‘Lift and shift’ existing workloads to their platforms. This replaces capital expenditures with operational expenditures, and reduces operational expenditures by consuming software as a service rather than self-managed software.
2. Attract developers to their platforms by reducing startup costs and increasing velocity by offering consumption-based versus capacity or allocation-based services. This enables developers to start small by using software as a service rather than self-managed software.
Every public cloud vendor has specific platform optimizations that can often lead to differentiated proprietary technologies and services. These proprietary APIs and services in turn can complicate running workloads on multiple public clouds or migrating workloads between them. Public clouds then represent the new homogeneous glasshouse for IT, locking in workloads with their proprietary APIs, technologies, and services.
Homogeneous platforms are not entirely a bad thing. A mature homogenous platform can offer a broad set of well-integrated capabilities. However, each homogeneous platform may come with a vendor lock-in cost. These costs can make application migrations to other public clouds expensive, could lack specific features that are required, or are not available where needed, such as where the data happens to be.
To circumvent these lock-ins and reduce dependency on one specific cloud vendor, customers often use an abstraction layer to make their workloads more portable across public clouds. Popular abstraction layers include RedHat OpenShift, VMWare, and Cloud Foundry, and each comes with their own vendor lock-ins. Abstraction layers make moving workloads between public clouds easier. For example, if you have 1,000 applications on OpenShift, you can move your entire OpenShift estate from AWS to Azure to Google Cloud Platform and back to AWS with relative ease. However, migrating 1,000 applications from OpenShift to any of the Kubernetes-offered services, such as AWS EKS, Azure AKS, or GCP GKE, is a more expensive migration.
Business How Tos and Recommendations
Public cloud “as a Service” is revolutionary because of its mode: consumption-based, self-service, and pay-as-you-go. Clients can scale applications up and down, offload infrastructure management, and rapidly consume innovative technologies immediately. Transparent prices give customers clarity on what they will pay and how costs compare between vendors. This same transparency enables customers to immediately consume without negotiations, physical provisioning, or procurement cycles.
About 35% of public cloud costs are wasted according to various reports, and billions are spent on unused capacity. [6]
We researched how companies could cut down on wasteful spending on unused capacity. Let’s say a customer is running an on premises VMWare infrastructure at only 70% utilization. If this customer wanted to move everything “as-is” to a public cloud IaaS where it runs at 70% utilizations, the customer’s Total Cost of Ownership will likely increase due to the capacity-based billing of IaaS [7] (you pay for 100% capacity, yet use only 70%). In order to see a benefit from the capacity-based billing models of public clouds, businesses must refactor their applications to take advantage of the potential economies of scale and elasticity of clouds. To get the best value, though, a business should choose the highest abstracted platform possible to take advantage of moving from a capacity-based billing model to a consumption-based billing model that corresponds to a company’s actual usage.
Overall, the best practice today is for a company to migrate IT to Application Frameworks and Databases as a Service to take advantage of consumption-based pricing, rather than operating their own middleware and databases on public cloud infrastructure. New advances in serverless computing platforms such as AWS Lambda or IBM Cloud Engine, enable customers to start small and rapidly scale when required. Serverless computing takes the pay-as-you-go model to the extreme by charging per compute unit millisecond.
Having an application that accrues cost only when actively used allows you to better align its cost to the value it generates.
In most markets, such as database management systems (DBMS), the SaaS delivery model is outgrowing the software delivery model [8]. IDC predicts the following markets will see software deployment shrink, and their “as a Service” presence grow:
● Content Analytics and Search Software
● End-user Query, Reporting, and Analysis
● Enterprise Performance Management Applications
Data Challenges
Applications store, use, and transform data. Consuming these applications as a Service means movement of data to the service provider. Businesses, however, often have data that lives in many locations. In a 2021 study by Forrester, 99% of respondents stated that their enterprise data is at least partially distributed.
In this same study, respondents confirmed that in the next few years their on-premises data estates will shrink while their data estates on public and virtual private clouds will grow.
Additionally, the study found that companies have to accept expensive trade offs due to the difficulty of moving data. Within the last year, cloud providers have begun to develop technology to address data challenges and concerns.
If data can’t be moved, technology that enables services to use the data anywhere “as a Service” allows customers to benefit from the cost-benefits of the “as a Service” model while retaining their data location. Businesses can for the first time use consumption-based models and services that move to where the data is rather than the data moving to the service.
Different Strategies
The hyperscalers (AWS, Azure, GCP) have prioritized expansion of their cloud service platforms to customer-on-premises before supporting competitive cloud service platforms. Vendors without a cloud service platform themselves, such as Snowflake, Mongo, Dremio, and others, quickly started supporting all three major cloud services providers, but continue to lack on-premises support. IBM, though, with its niche cloud service platform, has the most complete vision for offering services on-premises and supporting the three major hyperscalers.
Early adopters should choose carefully which limitations they can initially accept, whether those are features and functions or support of a specific cloud service provider or on-premises. What follows is a summary of the strategies hyperscalers and service vendors currently employ
AWS Outpost
AWS Outpost started as a hardware extension of AWS cloud to on-premises data centers. Initially, only a few AWS services were available on Outpost, but that list is growing. Outpost’s main limitation is that it is only available on-premises and does not support third-party clouds. In early 2021, AWS [9] announced Elastic Container Service (ECS) Anywhere and Elastic Kubernetes Service (EKS) Anywhere. Currently, ECS and EKS Anywhere focus on running container-based workloads, and no additional services such as databases have been announced to run on top of ECS and EKS Anywhere yet.
Microsoft Azure Stack
Microsoft Azure Stack, similar to Outpost, is a hardware extension of Azure Cloud to on-premises. It currently does not support any third party clouds. Microsoft Azure Arc is an extension of the Azure control plane (Azure Resource Manager) to on-premises and third party clouds. Arc does not bring its own Kubernetes, but can manage existing Kubernetes deployments such as OpenShift. Arc supports a limited set of database services on top of customer-provided Kubernetes. Arc Machine Learning has also been announced.
Google Anthos
Google Anthos focuses on application modernization by enabling customer-containerized workloads to run on Google Kubernetes Engine (GKE) anywhere. Anthos comes with GKE as a service, but can also control existing Kubernetes deployments such as OpenShift. Currently, the only service available other than Kubernetes is Anthos BigQuery Omni, which Google plans to offer on third party clouds.
IBM Satellite
IBM Satellite offers OpenShift as a Service anywhere, both on-premises and on third-party clouds. It allows customers to control OpenShift as a Service from a single pane of glass, addressing application modernization and container as a Service use cases. IBM also has a roadmap to bring its Cloud Paks, essentially the whole breadth of the IBM software portfolio from databases to middleware, to security and AI, as a Service to Satellite. IBM is unique in having strong support for all cloud vendors and on-premises solutions, along with a breadth of business services no other vendor has.
Service Vendors
Snowflake, Dremio, Instana, Mongo, and other service vendors lacking their own cloud typically offer their service on all three hyperscalers, but not on-premises. Some offer customer-managed software for use on-premises, but most focus on cloud as a service-only delivery model.
Summary
Anthos and EKS Anywhere seem to focus on application modernization, but currently lack services such as databases to help customers on their journey to as a Service anywhere. Azure Arc on database as a Service anywhere, but lacks application modernization capabilities. IBM has the most complete vision. It offers OpenShift anywhere, the most complete platform for application modernization, including all IBM Cloud Paks as a Service, as well as cloud services from the IBM cloud catalog such as IBM Cloud Databases.
Conclusion
The next decade will bring further fragmented enterprise data estates. Data movement challenges are likely to increase due to regulations and privacy laws. There is potential that some public cloud vendors may lock in their customers via costs such as egress charges. Enterprises will look to further their adoption of consumption-based, self-service, pay-as-you-go software models. This should help lower their costs and increase development velocity. The use of “as a Service” anywhere throughout the IT stack should increase efficiency by:
- Having workloads benefit from low-latency connectivity to co-located data without requiring additional software management.
- Processing sensitive or regulated data can be done in the data plane, next to the data. This allows the processing to happen without having to move the data.
- Rolling out workloads anywhere can be done faster, as the operational burden is with the service provider.
- Having a single pane of glass, from which workloads in all locations are managed from one management service for all locations.
- Administrative and governance policies across multiple locations can be managed from one management service for all locations.
Enterprises will increasingly expect their software vendors to offer these capabilities “as a service” anywhere their data is — which increasingly is everywhere.
The hyperscalers — AWS, Azure, GCP — will continue to focus on extending their cloud service to on-premises. Eventually, their focus will shift to offering their services on their competitor’s clouds. The hyperscalers lack of supporting their competitors platforms presents an opportunity for “as a Service anywhere” vendors without their own cloud, or a niche cloud, such as IBM, Snowflake, and others.
The ability to offer software as a service next to data will define how big of a market a vendor can address. Much data remains in on-premises data centers. IBM currently has the most complete vision offering software as a Service anywhere, including all three major hyperscalers and on-premises, addressing the biggest possible market for its software as a service anywhere.
What do you think? How will the various vendor strategies play out? What choices will customers make? Will they choose more homogenous platforms or heterogeneous abstraction layers? I would love to read your comments!
REFERENCES
[1] https://en.wikipedia.org/wiki/Write_once,_run_anywhere; https://en.wikipedia.org/wiki/Decentralized_computing ; https://en.wikipedia.org/wiki/Software_portability
[2] Arvind Krishna Interview, Think 2019, https://www.youtube.com/watch?v=hVQIrLpoyG4; Arvind Krishna Interview, Red Hat Summit 2019, https://www.youtube.com/watch?v=_3Gs_Ue5ujc&t=6s
[3] UNIX is a registered trademark of The Open Group
[4] Oracle, Java, and MySQL are registered trademarks of Oracle and/or its affiliates
[5] How to Manage and Optimize Costs of Public Cloud IaaS and PaaS; Gartner (G00465208); Meinardi, Clayton, Mar. 2020
[6] Why there is so much cloud waste; Dzone.com; Aman Juneja, Oct 2019, https://dzone.com/articles/why-is-cloud-waste-so-huge
[7] Predicts 2020: Better Management of Cloud Costs, Skills and Provider Dependence Will Enable Further Cloud Proliferation; Gartner (G00450736); Dec 2019
[8] The Future of the DBMS Market Is Cloud; Gartner (G00347472); Jun 2019
[9] ‘Go faster’ with core cloud services: AWS re:Invent 2020 roundup; 451 Research; Fellows, Rogers, et. al, Dec. 2020
Google demonstrates Anthos momentum; 451 Research; Fellows; Sept. 2020
How to Bring the Public Cloud On-Premises With AWS Outposts, Azure Stack and Google Anthos; Gartner (G00466935); Wright, Hewitt, et. at., Apr. 2020
Release It!: Design and Deploy Production-Ready Software; Michael Nygard
Breaking the fourth wall: New multicloud models bring the hyperscaler cloud to the on on-premises datacenter; 451 Research; Rogers, Atelsek, February 2021
Cloud Price Index: cloud-to-ground and cloud-around; 451 Research; Rogers, Atelsek, January 2021
Differences Between AWS Outposts, Google Anthos, Microsoft Azure Stack and Azure Arc for Hybrid Cloud; Gartner (G00720202); Sep 2020
