actionitem

Today three cloud-first architectures have been announced and delivered in beta or more. These are AWS Outposts, Microsoft Azure Stack, and Oracle Cloud-at-Customer.

I would expect other similar architectures to come from SaaS vendors, Alibaba, and specialist cloud operators.

we are further along with multi-cloud options than "True Hybrid" as described in this discussion. AWS position in the market makes them the one to watch, with the VMware partnership very interesting. Microsoft has apps the most places.

There are at least five vendors in this space: AWS, Microsoft, IBM, Oracle, and VMware. AWS is mostly in the future, pending Outposts GA, ditto VMware. In terms of "best," I'm not sure that any of these provide "identical" IaaS/PaaS across public and private. I'll punt

- I would say the combination of AWS and VMware are on track to provide that best hybrid cloud experience today. Azure stack is still maturing and Oracle is pretty far behind. Google / Nutanix may be a player one day but not right now.

VMware has many of the properties required for Hybrid cloud, and I would expect that they will offer and adopt these capabilities soon.

@dfloyer I would agree. Cloud at Customer on paper is the closest. I'm not qualified to comment on Azure Stack in depth - Outposts is currently a paper tiger and won't ship until next year - otherwise it's roll-your-own hybrid.

- I'm a broken record on this but hybrid would have been much further along if AWS bought Eucalyptus years ago instead of HP.

@ballen_clt Good analysis - I agree. Might add Pivot3 for security applications.

The largest amounts of data are likely to be found at the Edge. It is enormously expensive to bring this data to a centralized cloud. Lower cost to send code to decide what data should be analyzed and analyze it at the Edge, and send back small amount of data back.

Edge devices are increasingly running containerized workloads and exposing serverless interfaces. Within hybrid-cloud public/private mesh fabrics, edge devices and gateways may handle the bulk of ML inferencing functions and some training.

- Good point Peter. The edge is a macro-example of a hybrid app in many ways. Self-driving cars. Amazon Alexa... data moves from device (onprem / in-home) to cloud and insights come back directly or via software upgrades.

@dfloyer The point you made about apps effectively polling edge devices for data that can help with analytics, control, training, etc. is very important. How might that work?

@dfloyer are we talking microservices? functions? or something new? Still trying to wrap my head around the terminology since it isn't what I hear at the developer shows I've been to...

Smarter analytic functions in the cloud that understand latency and can automate inference code distribution to the Edge.

@ballen_clt An edge device or gateway might, in the extreme, functions as a sort of "mini-private cloud" in a larger hybrid-cloud hub-and-spoke architecture hubbed around a public cloud.

@dfloyer - I gently disagree. Many edge devices will not have the computing power or time to effectively process all the data they collect. Think sensors on jet engine turbines. Much of that data will go to a central repo / cloud until we figure out what to do with it.

@ballen_clt - We don't yet know all the questions we want to ask of data collected at the edge. Much of the power of edge data lies in aggregating across larger solution sets - which has to happen centrally for now

@ballen_clt there are multiple edges. Some edge devices can process (cars, AWS Snowball), some will need a device on cell tower (still close).

Kubernetes is not a magic layer; it will help find the commonalities and minimize the extra differences that didn’t need to be there. Here's clip from @jbeda of Heptio How Kubernetes fits into the multi-cloud discussion

Open source in general is a rich source of innovation for all cloud environments. Kubernetes has become a de factor standard for some groups of applications. Other inventions required for other groups such as High-value systems of record.

The evolution of Kubernetes will under CNCF governance, and vendor commercialization, will hopefully converge on a common vendor-neutral container/orchestration stack. As that happens, hybrid-cloud Kubernetes deploys will be less painful, hence lower cost/risk.

- Kubernetes will encourage folks to try and fail faster (in a good way). There will be less fear about vendor lock-in.

High value applications, and applications with large amount of distributed data will both get value from hybrid cloud.
An example of a high-value application is adding real-time data from analytic/AI systems or systems of engagement - e.g., fraud detection.

- Hybrid only makes sense if you don't have to pull large amounts of data down from a public provider. The egress costs quickly erode any other benefits. Apps where you upload lots of data, process in public and ship "insights" back to private venues more sense.

Hybrid application architectures, as I've defined them, best serve apps that can benefit from serverless development abstractions and containerized cloud-native workloads in a virtualized multicloud fabric. That well-suited many analytics pipeline workloads.

@ballen_clt - I give you lots of data and you give me back a simple answer.

@ballen_clt 42 ;)

Hybrid cloud data architectures, however, are a different creature. They suit applications in which some data must stay on-prem but other data/workloads may run and be scaled out in public clouds. Many customer-facing multichannel marketing apps fit that description.

@ballen_clt A central, practical question for packaging of AWS Outpost. Most likely, though, egress charges will be moderate within hybrid cloud (AWS-to-AWS) implementations.

Reason for Hybrid Cloud & Hybrid Applications - the three laws of cloud: Physics – real-time requires low latency, requires compute to be close to data. Economics – high cost/latency for moving large amounts of data  Move code to Data
Politics – legal, compliance &

hybrid cloud is spanning public and private cloud. Managing both the stuff I run, and the stuff someone else runs. Today most applications can run in many places, but don't span between geographic locations.

The notion of "trustworthy" hybrid cloud nodes is especially important: Enterprises want to be able to target work wherever it is optimal and not have to worry about verifying targets.

A hybrid cloud is a computing environment that integrates at least one private cloud and at least one public cloud behind a common control plane; supports unified security, monitoring, and governance; & enables flexible movement of apps, data, and workloads amongst.

Reason for Hybrid Cloud & Hybrid Applications - the three laws of cloud:
Physics – real-time requires low latency, requires compute to be close to data
Economics – high cost/latency for moving large amounts of data & Move code to Data
Politics – legal, compliance

Multi-cloud frameworks connect public and private clouds together;
Hybrid Clouds offer much tighter integration across nodes.

Another subtlety is RESTful or not: Hybrid cloud should better support stateful interactions, whereas anyone deploying multicloud apps not using RESTful techniques does so at their own peril.

A hybrid application is built in a development environment that enables composition of business logic in two or more abstractions (eg VM, containers, functions) that are executed in disparate runtimes & back-ends.

Hybrid Clouds offer much tighter integration across nodes.
Migrate applications/code from any node to any other node without conversion, migration, recompiling or testing
Same Control Plane for all Nodes, Services-led
Key Enabler for high-value Hybrid applications

HYBRID APPLICATION tech will accelerate changes to HIGH-VALUE TRADITIONAL APPLICATIONS by showing the development/maintenance advantages of refactoring trad'l apps into loosely coupled microservices.

The potential improvement of existing mission critical systems-of-record will enable greater capabilities to automate business processes. A key aspect of this is enabling systems of engagement and analytic/AI systems to enrich systems of record.

SaaS vendors will be a major contributor to hybrid systems and hybrid applications. CIOs are wanting their data to remain on-premises, so it can be accessed by other applications, and so they can be more certain of confidentiality.

SaaS vendors influence the direction of HYBRID APPLICATION tech/adoption through adoption of serverless and containerization into their delivery of on-demand functionality and through their connectors to IaaS/PaaS offerings that do likewise.

SaaS appliances using their own PaaS are already in the marketplace.

A low-latency control plane is an essential building block of hybrid clouds and hybrid applications. These are likely to be separate and specific to the hybrid cloud. There will also be requirements to interface with multi-clouds, at higher latencies.

"Independent" is the wrong concept. The multi/hybrid cloud control plane can be built in a vendor-agnostic way, so that it can be used to manage, monitor, and support DevOps over various combinations of public, private, and edge clouds.

The control plane for hybrid cloud systems should incorporate platform/vendor agnostic infrastructure-as-code interfaces to drive DevOps of VMs, containers, and serverless functions across heterogeneous cloud back-ends.