VMWare is doing you a favor!

Cloud solutions chosen in the past for different reasons do not meet the needs of the future. But the easiest choice is to stay with what you have. VMWare is doing us a favor by forcing a change at the start of a new wave of market needs and opportunities.

Moving forward, what do we need, and what should we remove?

What criteria should we use to benchmark our ability to meet the future with a suitable operational model, economics, and capabilities?

This is what we shall discuss in this article

Introduction

We need to make new cloud choices for various reasons.

1. Incumbent changes of ownership
2. New market opportunities
3. Multiple cloud provider supply side
4. Financial realities (cloud done wrong is expensive)

When we made decisions before, it was easier; there were fewer choices.

When we refresh these decisions, we should realize how important they are. In many ways, your cloud choice defines the business you can run. How fast can it move and change direction, and how expensive must operations be to get both?

What is cloud? The fundamental role of cloud computing is to abstract operations from physical infrastructure using software representations of useful programming interfaces and services, enabling more flexible, scalable, and cost-effective IT resource management.

Many of us are changing our approach to cloud.

This article will show

1. that you are not alone.
2. AI is forcing computing to move to the data, a reverse motion from what has happened.
3. Managing one cloud is no longer the problem.
4. There is a need to revisit what "cloud" is.

We want to start with understanding why this is happening now.

Incumbent Changes of Ownership

Let's be transparent. We all know we are speaking about VMWare. Broadcom is doing its customers a favor by forcing many to reconsider how to solve their cloud computing needs. VMWare was designed in the 2000s to hide physical server reality, IT manual bottlenecks, physical server installation lead times, and under-utilized server bloat.

It improved IT and server installation waiting times while creating the problem of virtual machine bloat, a 10x problem due to removing the physical constraints and barriers. I am unaware of any IT organization that reduced size after adoption.

"This is the central deceit of cloud marketing, that it’s all going to be so much easier that you hardly need anyone to operate it. I’ve never seen it. Not at 37signals, not from anyone else running large internet applications. The cloud has some advantages, but it’s typically not in a reduced operations headcount." [ David Heinemeier Hansson ref]

VMware was not designed for the world we are moving to, but like all incumbent solutions, it has become very integrated in the past. Without the forcing function now being presented, most organizations would not consider removing such a deeply embedded incumbent; it feels easier to extend instead what you have to cope with future requirements. But incumbents were not designed for this reality. They are made for centralized, not distributed, operations. The cost of their control plane overhead can be hidden in the centralized operation, but when this is required in computing-constrained locations, the cost overhead becomes explicit and expensive. A highly distributed operation is very different from a centralized operation.

This leads to the market reality insight.

New market opportunities

AI will change the landscape of all digital architectures on the application and supporting infrastructure levels. Ignore emotions of hype cycles and questions of AI bubbles. In the hyper-connected world we now live in, immediate access to data is essential both for customer experiences and internal operational efficiency; having the ability for machines to interpret that data and make decisions in the moment transforms businesses from guessing to immediately knowing and being able to take granular immediate actions 24/7.

Computing capability has to move to the data, which diametrically opposes the trend from the last 20 years of moving the data to the compute. This changes the design center of all digital operations optimized for centralized operations.

Any cloud solution must be designed for

1. highly distributed operations,
2. low overhead per location,
3. high volume parallel automations.

This translates into fast, agile, and cheap business outcomes.

To speak in AI terms, if we were designing a model to maximize a goal, the goal of cloud computing would be Cost/(Speed*Jobs).

Put another way, the goal is to have the smallest cost in the fastest time for all jobs needed.

Using technology that is not designed for this will kill the business. Any compromise will lead to unsustainable cost structures and a lack of ability to cope with change. Many evils can be hidden in the scale of centralized operations. Running distributed operations can be likened to being outside naked with nowhere to hide.

Multiple cloud provider supply side

All large businesses have multiple cloud providers. In most companies, clouds are selected on a department basis. When companies buy other businesses, they inherit the cloud choices of those other businesses. A multi-cloud system is inevitable, even if a single cloud is the desired north star in strategy documents.

Success in the cloud is not about having one or many clouds; it is about having one cloud-native operation that meets the business objectives over as many clouds as required.

Therefore, the " cloud " selection is more about selecting the wanted "cloud operation" that can operate any infrastructure, be it off-premise, on-premise, proxied by another cloud, and/or bare metal servers.

The defacto interface for cloud-independent management is kubernetes (k8s).

Kubernetes is open source. Why not take that and run with it?

1. Kubernetes is very complicated to do well.
2. It focuses on container orchestration rather than operational or application life-cycle management.
3. It offers consistent interfaces but does not offer any automation.

When choosing a kubernetes-based implementation from a vendor, it is essential to require

1. zero modification from the k8s open source release
2. support of highly distributed wide area k8s
3. lowest possible overhead of the cloud control channel
4. day 0, 1, 2 operational automation

You cannot choose the number of cloud providers you will have to manage.

You can choose how you manage them. As one homogeneous operation, irrespective of whether it is for your internal offering development and/or to offer abstracted cloud capabilities to others.

Financial Realities

Once more, let us revisit what cloud is.

Cloud is the most effective digital operation, measured by speed, agility, and cost.

Many companies' cloud adoption has resulted in the opposite: an increase in direct costs and no increase in agility. When more clouds need to be managed, increasing operational complexity leads to increased direct costs once more.

The early adopters of the public cloud sang its praises, and their reasons were initially simple: the ability to move quickly and scale seamlessly based on actual growth rather than predicted success.

However, once success is achieved and demand curves become predictable, the public cloud quickly becomes prohibitively expensive. Because of the hard-coded dependencies in the proprietary management choices made and the cost of moving data, lock-in is achieved as growth and associated costs continue to grow as part of the success.

Public clouds are not a public service!

For just 2023, AWS's operating income was $24.6 billion, and sales increased 13% compared to the previous year [ref].

As Partha Seetala, President of Rakuten Cloud states

Failure in public cloud is really cheap, success is really expensive!

This is a publicly stated problem for digital-native companies born in the cloud, as Snap's IPO disclosure in its S1 registration shows.

Some internet companies have repatriated their workloads to private data centers. David Heinemeier Hansson, pioneer of Ruby on Rails - The Rails Foundation and CTO, Co-Founder of 37signals (makers of BASECAMP) has transparently documented the learnings made, and the remarkable savings achieved.

Managing multiple clouds is inevitable. Managing them to secure the highest speed, lowest overhead, and minimized cost is a business capability built on the right cloud-native operations.

Choosing which infrastructure to run workloads should not be a static design decision but instead be based on location requirements, traffic patterns, CPU utilization, performance requirements, provider economics, and stability.

I covered this in this introductory post from two years ago. When I say "telecom cloud," it is the same for any industry or company. I wrote for telecom because the analyst asking me questions was a telecom analyst.

The fact that workloads might need to move in the future means the ability to move applications, workloads, and data should always be automated and controlled by the business running cloud-native operations. The industrialization of operations is at the heart of financial success and is the real driver for cloud adoption.

A good cloud-native operation drives the industrialization life-cycle of continuous improvement; see below.

In conclusion

Cloud is not something you buy; it is something you do. This is why simply buying a public cloud and expecting different outcomes is no different from buying a private cloud and expecting the same. Both end in missed opportunity at best and expensive learnings at worst.

Success comes from empowering any internal team to become as capable as possible as quickly as possible with the right generational approach to cloud management software and operational tooling.

Rakuten Cloud consideration

We have already had to do this at a large distributed scale in?Rakuten Cloud?(previously Robin Networks),?Rakuten Symphony,?and Rakuten Mobile, Inc. The traditional method of building and operating the network was impossible; it needed to be faster and cheaper.

We quickly realized that we were not building mobile networks anymore but rather highly distributed, high-performance networked data centers that run local software with high real-time performance requirements. However, we cannot afford large overhead (tens of thousands of locations) or manual intervention. Traditional solutions could not even upgrade the software across all locations before it needed to be upgraded once more.

The application requirements demanded line-rate networking without jitter or delay (at the microsecond level), and input/output (I/O) performance had to match this high standard with ultra-low latency and high throughput.

As we move to the data-driven real-time AI world, all applications will start to have similar requirements and distribution. If a machine is delivering your business, the machine cannot take a break. As all businesses become real-time service providers of one type or another, all businesses will need the same level of distribution and guarantees.

This is the future technology world we are all going to live in. When we needed a solution, no existing one was available, so we acquired and built our own.

To be a future digital leader requires you to be a future cloud operational leader.

This is why this decision is so important and so timely. Worry less about the industry and consider more about the capabilities needed. Giant leaps in capability tend to come from adjacencies and cross-pollination of disciplines, not from homogeneous suppliers.

If you would like to learn more about Rakuten Cloud, you can visit here.

If you want more articles like this, subscribe to the Not Normal Newsletter.