2018 — Tech Predictions: Cloud, Containers, AI, and more
2018 welcomed us with some of the scariest security vulnerabilities of all time: #Meltdown and #Spectre. As with all end of years, we a saw number of predictions for the new year as well, with some that I found interesting (ML Trends, TNW, CTO Advisor, etc). I am listing down my predictions for this year, primarily towards challenging my own understanding of the industry trends. I request you to challenge my predictions either now and/ or at the end of the year.
- Year of Managed Services/ Managed Service Providers
- Year when Vendor Lock-in is better understood and embraced
- Year of Serverless, only too soon
- Year when Cloud started moving to the Edge
- Year when Cloud-Native is no longer just about Containers
- Year when Kubernetes became next OpenStack
- Year of major consolidation in the cloud-native eco-system
- Yet another year of IoT, only not to be
- Year of GxINFE (opposite of GxFY)
- Year when BTC crashed, Long live Crypto!
- Year of Managed Services/ Managed Service Providers
We are witnessing an increasing trend of cloud adoption through managed service providers across multiple cloud service providers/ platforms — I had earlier talked about how I saw customer empathy as the motivation for a slew of managed service announced by AWS at re:Invent last November. I notice the same trend with other cloud service providers (for example, GCP offering Cloud AutoML, Azure offering Managed Kubernetes) as well. AWS started offering its own managed services from 2016; Microsoft always had encouraged partnerships with MSPs with Azure being no exception; many managed service providers (such as Rackspace) now support multiple cloud platforms.
Also, consider the levels of adoption maturity of customers across regions. Even with optimistic estimates, only about one-fourth of the organizations have mature or optimized cloud adoption strategies (IDC observes less than 10% use optimized practices and in some regions, far lesser). With all these, it is easy to see that that managed service providers will play a pivotal role in helping enterprises migrate to the cloud.
I am also willing to bet that the revenue through managed services (across all providers) will surpass revenue through underlying infrastructure (raw compute, storage, functions, etc).
Currently, no cloud service provider makes the revenue numbers available at this level of granularity. But consider this — if you are an enterprise leveraging cloud through managed services, which component of your bills is higher: resources or services?
2. Year when Vendor Lock-in is better understood and embraced
Vendor Lock-in has been the most cited reason for not limiting to one cloud service provider or a platform. This was beaten to death in the OpenStack eco-system during the infamous distro-wars (ultimately it didn’t matter as far as OpenStack distros were concerned). This is also becoming a primary driving factor in the Kubernetes eco-system — with all cloud service providers now supporting Kubernetes, a lock-in free nirvana is being dreamed.
However, lock-ins happen in different ways and certain lock-ins are inevitable. Rather, are well understood and embraced. With current business practices, application development abstractions and technology options, there is no way to achieve a lock-in free nirvana (see my tongue-in-cheek report “How to avoid vendor lock-in”).
Does that mean to put all your eggs in a single basket? No! Select the right provider/ platform for your workload — based on your business & technology needs. Don’t let ‘no vendor lock-in’ influence your decision making, which will only have you locked you into otherwise avoidable CapEx, abstraction, operational debt or tech debt. Multi-cloud is the reality — leverage it to the fullest. As with everything in life, there is no single solution fits all.
In this context, I expect 2018 to be the year vendor lock-in is well understood and embraced. Rather, I hope so :)
3. Year of Serverless, only too soon
I am as excited about Serverless as any other analyst covering this space (but not more than swardley I guess), but 2018 will not be the year of Serverless. Not 2019 either. Not until 2023 at the least.
Yes, AWS Cloud9 IDE supports developing serverless applications; yes, TCO of serverless is likely to be lesser than VMs; yes, there is even an open source FaaS platform on Kubernetes (Fission); and yes, Serverless will fundamentally change how we code. But not in the near future.
Consider varying levels of customer maturity and the need for managed services. Consider vast majority of enterprise workloads that have not migrated to cloud yet (best case estimates place less than 15% have been migrated to the cloud). Consider the path cloud-native eco-system took from totally stateless workloads to semi-stateful workloads to stateful workloads. Consider how cloud-native is not only about tech, but also processes and culture. Consider Data Gravity. Consider training. Consider questions like ‘if I run a FaaS platform on my private cloud, am I doing Serverless’. Consider all these and where Serverless can take one to.
You will realize how far we are from the year of Serverless.
4. Year when Cloud started moving to the Edge
In terms of maturity, Edge computing is still at its incipient stages. Definition of the Edge varies depending on whom do you talk to — last mile connectivity, mobile device registration, cloudlets, AR, or more. The use cases relevant to the edge are those that involve low latency data analysis/ processing that can’t wait for a roundtrip to the cloud.
2017 saw interesting developments around pushing cloud to the edge — AWS announced the preview of DeepLens, OpenStack community rallied together on using OpenStack for edge computing, Microsoft announced the preview of Azure IoT Edge, etc.
2018 will be the year when the top 3 cloud service providers will have at least one compelling edge computing value proposition. I also expect the Kubernetes community (along with or independent of OpenStack community) to work on enabling edge computing use cases this year.
Please note that this is not going to the definitive year we call ‘year when the cloud moved to the edge’. While there appears to be a good understanding of the use cases that need cloud-like capabilities at the edge, definitions of this ‘cloud-like’ edge are far from being clear. For example, AWS Greengrass brings Lambda capabilities locally to the edge device/ IoT device, but those Lambda functions need to be authored in the cloud. Yes, OpenStack on an edge device has a smaller footprint, is it minimal enough? We don’t have the definition of ‘necessary and sufficient’ edge cloud yet.
I also expect to see discussions around models to monetize such edge devices emerge this year.
- Year when Cloud-Native is no longer just about Containers
The term ‘Cloud-Native’ is no longer only about a category of applications - it has evolved from the definition of applications to how businesses run applications to software architecture to design patterns to including everything - team, technology, and culture. It is also interesting to see what started as something standardized based on containers, has expanded to include higher levels of abstraction as well: Serverless/ Functions as a Service. And it might extend further to include abstractions to come in future too.
Even if we consider the level of abstraction standardized on containers, cloud-native applications have become complex enough to include aspects more than orchestration — logging, monitoring, visualization, billing, etc. As they get more complex and production-grade love, conversations would move past packaging and orchestration. And it will no longer be just about containers.
I see this as a natural phenomenon — as tools and technologies develop, use cases leveraging them develop too. And as these use cases get more complex, newer tools/ technologies develop. In this positive feedback loop, newer abstractions might develop, but operating principles may remain the same.
2. Year when Kubernetes became next OpenStack
Depending on who you are — you might nod in violent agreement to this or brush it off or smirk or even start hating me. But, Kubernetes is becoming next OpenStack, in ways both good and bad. Please note that they are distinct products, with unique advantages/ disadvantages, characteristics, and places in tech-history. They will never be the same, but my point is one is increasingly becoming like other.
Kubernetes is mature to the point that the community is beginning to realize that it is boring. So is OpenStack (hey, I called it first:)). A stable, mature product is always good for customers. Need to say more?
Kubernetes had seen adoptions since its early days and is only increasing. OpenStack’s adoptions haven’t flatlined either. They will continue to have respective successful user bases.
It is not a hyperbole to state that both projects owe their success and momentum to their communities. While the OpenStack community was one of the fastest growing one, Kubernetes community is one of the most inclusive communities I have seen. I am fortunate to be considered part of both communities and I am thankful for that.
Complexity, Ubiquity, and Customer Maturity
Both Kubernetes and OpenStack have large scale use cases to showcase, but the reality is that the amount of work and expertise required to deploy and operate OpenStack clouds/ Kubernetes clusters of any decent size is not trivial (Kubernetes has an edge over OpenStack on the ease of deployment). Combine this with the levels of customer adoption and technology maturity - for a vast majority of customers, they are still scary buzz words. It is also to be noted that while OpenStack and Kubernetes solve specific set of problems relatively well, there is a lot more than just the building blocks (compute, storage and networking) or VM/ container orchestration.
You will soon realize that they can only be parts of the puzzle called enterprise IT, not the whole of it. You will also soon realize that they are the components of enterprise IT that can be commoditized easiest.
OpenStack is the Linux for private cloud.
Kubernetes is the Linux for containers (sorta).
Ubiquitous; Essential; but not shiny pieces anymore.
Not Built Here
By not being open to learning from other’s experiences, the Kubernetes community appears to be repeating some of the mistakes that the OpenStack community did.
Both OpenStack and Kubernetes communities seem impacted by ‘‘not build here’’ syndrome. Remember the infamous “AWS API compatibility or not” debates early on the OpenStack eco-system? Kubernetes appears to be treading the same path (RBAC, QOS, trying to be more than orchestration layer etc.). Both communities also seem to be driven by a lot of ‘us vs them’ sentiments. For the OpenStack community, ‘them’ was AWS; for the Kubernetes community, it is OpenStack.
Kool-Aid, until Next Hype
OpenStack was the coolest kid on the block. Everyone drank OpenStack kool-aid. Until Docker & Kubernetes came.
Kubernetes is the coolest kid now, everyone drinks/ will drink Kubernetes kool-aid now. Until Serverless becomes one.
When you see this image, if you get reminded of the days when OpenStack vendors bragged about their code contribution metrics, # of PTLs they employ etc., I think I have made my point.
Clearly, 2018 will be the year Kubernetes becomes next OpenStack.
3. Year of major consolidation in the cloud-native eco-system
Ok, if you are a startup or a vendor participating in the container eco-system, brave yourself for a major consolidation this year.
Consolidation in any technology eco-system is inevitable; only the pace at which it happens varies. There is only one way to make money from open source software and only one company has been able to do it at large scale so far. I expect the consolidation to be rapid in this eco-system. It has already started to happen in the in the cloud-native/ container eco-system.
If I were a startup in this eco-system, I would be worried -
if I don’t have a path to 7 digit revenue or a roster of paying enterprise customers
if I had raised more than $100M in funding
if I don’t have impressive engineering talent
if I don’t have established services/ support credibility
If you are a startup with a reasonable revenue and impressive talent, you will make a good acquisition target. You have a window of about 12 months to cash in. If you are approached with offers of
$25M+, consider it.
$50M+, seriously consider it.
$100M+, TAKE IT NOW.
$250M+, how cute! Please wake up from your slumber…
- Year of continued exponential growth of AI-aaS offerings
All big 3 public cloud service providers (AWS, Azure, and GCP) now offer various AI-aaS capabilities, including Image Recognition, Video Recognition, Language Translation, Transcription etc. IBM Watson also offers a slew of similar services. They also offer features/ tools to derive value out of ML sooner.
Last year, I had postulated that
Cognitive computing capabilities available as a service will double approximately every year
This a natural extrapolation of Moore’s law. And don’t forget all those GPU goodness that’s available to developers now (AWS, Azure, and GCP). I expect this trend to continue this year, with possibly new players (either large providers such as Alibaba or Huawei, or niche players) entering into the foray.
2. Year that would pave way for next AI winter
It is my reading that this cycle of AI will not deliver meet the ultimate promise of AI and need to wait until the next cycle of AI to achieve equivalence of IQ and EQ of humans.
For all the impressive advancements we have witnessed so far, disappointments galore in some simple, common use cases (for example, voice-controlled home automation). If you look at the techniques involved, the models and theories have remained almost the same for the past 20 years or so. It is mostly the advancements in computing power that have enabled recent achievements in advancing AI capabilities. Above all, we also have unanswered questions on ethics, bias, and moral issues.
With all these, one wonders if AI will be able to keep up with its promise?
What we need is not just faster computing, but different techniques, models, and technologies. We will soon reach the upper limits of Moore’s law, even with GPUs and reach a point where merely performing at speed is going to be enough.
We will need new models and computing paradigms — I am betting that only with sufficient advancements in Quantum Computing we will be able to meet the ultimate promise.
I expect discussions around these inadequacies of current advancements in AI to strengthen this year. I also expect that next AI winter is on the horizon (2–3 years from now) and such discussions will starting paving way for that
- Yet another year of IoT, only not to be
We have seen various projections on connected devices/ IoT devices (see this for example), but they are nowhere close to the reality. Why?
Is it the lack of management tools/ services? No
We have a variety of IoT management tools/ services at our disposal now. We also have IoT Management tools/ services offered by cloud service providers (AWS, Azure, GCP, and IBM), open source platforms (OpenIoT, Edgex Foundry), and industrial IoT device management platforms (GE, Hitachi, etc.). So, it is not the lack of management tools or services that is blocking from realizing the full potential of IoT devices.
Is it the lack of hardware capabilities? Mostly No
With devices such as AWS SnowBall Edge, smart home automation devices, industrial IoT devices, and plenty of such devices, it is not the lack of hardware capabilities that seems to be the blocker.
Is the user experience not convincing enough? Maybe
There is some truth to this. For example, if you want to improve your home automation, you are posed with too many options/ standards and vendors, each trying to get you to limited to their devices. You will also encounter different methods of setting up/ configuration which may not be very intuitive. While all such devices are vast improvements from their predecessors, like the good old Honeywell thermostat, they are still lacking the sophistication and ease.
Is the market fragmented? Yes
Yes. When you use an IoT device, there are multiple players involved in providing you with that experience: hardware/ device manufacturers, software/ services providers, platform vendors, connectivity providers, application developers, and retailers, each with possibly more than one choice. The market is clearly fragmented.
I draw an analogy to the smartphone market here — in this case, each consumer will own more than one device. More than innovations in mobile hardware, software, and supply chain, deeper penetration of mobile phones were possible due to the success of last mile connectivity: cell phone carriers were able to provide that through successive spectrums.
Without such last mile connectivity, IoT devices will only be cool hobby projects or expensive industrial projects, without realizing their fullest potential.
This year is not going to be the year that happens. We may need to wait for 2–3 years.
2. Year of GxINFE (opposite of GxFY)
What is GxIFY? Google X For You, as in Google Infrastructure For You, Google Security For You, etc.
One of my peeves of 2017 was how the term SRE was abused. It was touted by Google — how it benefitted them, how anyone can follow SRE principles, and why everyone should do that.
My other peeve: how Kubernetes was in general marketed as ‘Google runs infrastructure with this; now you can also run your infrastructure with this’.
In general, GxFY misses an understanding of who the customer is and what their capabilities are. Just because something worked for Google doesn’t guarantee it will work for all. It may work if the customer is ‘Googley’ enough :)
I expect this year to bring an understanding of what GxFY is, and what one needs to do to find more success with GxFY, and a realization is that Google x Is Not For Everyone: GxINFE. And that is OK.
3. Year when BTC crashed, Long live Crypto!
This is out of my comfort zone, but I am nevertheless going to put my stake.
This is the year BTC will fizzle out, but it won’t be an overnight phenomenon. I see various factors contributing to this:
Power/ Compute Requirements
Mining Bitcoin is becoming insatiably power consuming. It is becoming more difficult and expensive to mine a bitcoin. If this trend continues, it would arrive at a state where there is no more compute power available to mine a bitcoin (at rates that gives you return). Or you might need custom ASICs or special purpose hardware to do so.
If you take the analogy of mining for minerals, you will arrive at a stage where it is no longer profitable to mine.
Driven by Greed
Bitcoin appears to be at a stage where it is no longer catering to its original principles, but catering to greed based upon its limited availability. In human history, anything driven by greed has crashed and will continue doing so. It is then governments get in to regulate (and of course, try to get a slice of it :)).
Cryptocurrencies are virtually unregulated now — one enters ICO at one’s own risk. This is the time for regulations to be put in place to improve stability and market confidence. It is interesting to see some efforts around this (US, South Korea, India, etc). While I don’t expect final regulations to be put in place by this year, I expect good progress on how these regulations might look like.
Cryptocurrencies are here to stay and this year is going to be vital in establishing required regulations and guidelines.
Here is to an exciting 2018!