One Cloud to rule them all? (Part 3 of 3)

Written by
Etienne Gadient

A group of passionate Technologists, Consultants, and Trusted Disruptors focused on the maelstrom that is Cloud services and the IT industry. Get decisions close to the data, be disruptive, and design for cloud in scale. You've been warned.

One Cloud to rule them all? (Part 3 of 3)

Written by
Etienne Gadient

One Cloud to rule them all? (Part 3 of 3)

Written by
Etienne Gadient

Two weeks ago, AboveTheCloud (ATC) started a multi-part series to explore the strengths and weaknesses of best integrated vs. best of breedIn part 1 of the series, we explored the foundations of best of breed and best integrated as architecture principles and compared the industry major players, including Amazon Web Services, Microsoft Azure and Google Cloud.   In part 2 of the series, we explored the benefits of focusing on best integrated, also known as a "single cloud" focus. 

Today, we will focus on the benefits and use cases of Best of Breed services across a portfolio of multiple public cloud providers.  

A Brief Sidenote on ATC's Independence

In full disclosure, I am personally certified as a Google Professional Solutions Architect.   ATC, as an entity, maintains independence from all commercial cloud providers and does not receive financial or promotional compensation from any entity to ensure its position of unbiased opinions.   As a result, it is important to note that EACH cloud has best of breed features that uniquely differentiate themselves.  To illustrate this concept, I am going to use a Google Cloud capability as just one of many examples of differentiation. 

Go Big or Go Home

One specific example is a best of breed big data service. Google Cloud Spanner is a unique set of relational database services, that simultaneously support strong sequential consistency (ACID) characteristics and massive horizontal scale, typically mutually exclusive requirements.  In database terms, massively scalable distributed databases (i.e. NoSQL) will eventually become consistent, however there is a limitation to whether you can guarantee transactional consistency across multiple distributed databases at any specific point in time.   Relational databases are known for having strong ACID characteristics, in other words, can typically be guaranteed to be consistent and accurate in any moment.  This may sound like a minor issue, however, if you are the NYSE or a major bank, it is vital that your bank account value database entries are all consistent when a client wants to withdrawal money.   Other database types, such as a product lookup catalog do not require significant ACID characteristics and do not require the same rigor.   ACID characteristics are why relational (i.e. SQL) databases have always been the gold standard of trust, while non-relational databases (i.e. NoSQL) are the gold standard for huge datasets and massive horizontal scale.  In the public cloud, if you have a requirement for a massively scalable database with strong ACID characteristics, the only real choice is Google Cloud Spanner.  Supported by Google's Colossus file system and some clever use of sharding, Cloud Spanner supports hundreds of Petabytes or more while also maintaining strong ACID characteristics.  Nothing else can solve for this use case and Best of Breed should be the choice to meet the requirement.

So How Hard Can It Be? 

Embracing a multi-cloud portfolio is not something to be taken lightly.  Once a use case drives a best of breed requirement, the implications of the choice become important.  Taking our cloud spanner example deeper, the choice to deploy this massive transactional database into a new cloud provider means that all related and/or dependent workflow applications with high volume, low latency affinities to that database now must also be operated out of that new cloud provider, requiring monitoring, backup, and operational foundational services.  

It gets covered in part 2, policy enforcement as an example. The three largest clouds can do similar things in very different ways.  Another example of this is network transport.  When designing multi-regional applications for Amazon's and Azure's networks, network traffic traverses the internet with a variable latency and the potential for multiple paths to be taken.  In Google's Cloud, traffic transiting between Google regions traverses Google Cloud's semi-private (well, its private to the many Google Cloud semi-private) backbone with less variability in routes and typically lower latency.  In fact, Google keeps traffic in its semi-private network until it reaches the closest point of presence for the end user.  This is a huge differentiator for application workflows that span across multiple regions that are latency sensitive, but changes many of the implications for designing applications across regions.  

Sound complicated?  It is. 

How Far Does This Rabbit Hole go? 

Best of Breed also requires that you have access to talent who can translate how to do similar functions with this new cloud...and have defined all of these functions specific to this cloud: 

  • Monitoring tools and standards
  • Capacity Planning
  • Service Catalog
  • Cost and Usage Optimization
  • Data Protection
  • Security and Policy Enforcement Methods
  • Service resiliency methods
  • Billing and Cost Allocation methods and tools
  • Skills and Context

But That Is A Lot of Work...

Opening up a new cloud provider is similar to opening up a second front in a war.  It takes investment and commitment along with the ability to support both simultaneously.  In other words, does the financial or capability benefit outweigh the additional cost and complexity of managing both? For large enterprises, a large percentage find it is.  How can that be?

  • Best of breed capabilities can be made available within each cloud
  • Improved service resiliency from a provider outage
  • Can take advantage of unit cost opportunities between Cloud Service Providers
  • No single blast zone 
  • Can be more cost effective in larger scale
  • A critical business requirement that can only be solved by a best of breed capability

Why Are We Doing This Again? 

One company leveraged a multi-cloud framework to break the control their outsourcing providers had on them.  By decoupling hosting, solution design and operations management into different disciplines, they were able to renegotiate same or better service level agreements while reducing the cost of IT by $25 Million a year.  How did they achieve this?  They developed foundational services in one cloud provider and built a critical mass of IaaS and PaaS services and then took their cloud center of excellence team and diversified roles to specialize in each CSP supported. Using this foundation, existing and successful solution patterns were translated to each CSP.  As a result of their scale, they had specialists per CSP aligning the portfolio to take advantage of the best of breed capabilities among the CSPs.   Leveraging the restructuring of their outsourcing provider portfolio (and weeding out players who couldn't adapt) they were able to reduce the overall cost of IT even with a small increase in unit cost for the IaaS resources.  

The Wisdom of Multi-Cloud

If an IT organization is big enough and mature enough to support multi-cloud diversity, there are many financial, technological and operational reasons to consider a multi-cloud portfolio of services. Another strong example of multi-cloud benefits involved Netflix.  Netflix has always been a "cloud darling" with much publicity and attention going to the relationship between AWS and Netflix.   Many years ago, AWS had an outage that caused a global impact.  During this outage, people noticed that while AWS users were impacted, Netflix users were not, despite Netflix's very public model of deploying all related services into the public cloud.  The mystery was solved when a few resourceful journalists were able to confirm that Netflix ran a percentage of their movie streaming and account management services out of Microsoft Azure for diversity.  To achieve this, they had separate development teams and code bases for each CSP.  Yet, they had limited their blast zone to still provide a high quality of service despite a major outage of their primary CSP.      

We hope you have enjoyed our three-part series as much as I have creating it.  In the end, the choice will be based upon your use cases, scale, and maturity level.  Both models are viable, and ultimately depend on the organizational characteristics driving the requirements.  When the proper planning, time and investment is made, Best of Breed can give you the most capabilities while also having a ton of agility to offer.

Part 1: Single Cloud vs Multi-Cloud trends and comparing the clouds...

Part 2: The benefits of a Single Cloud Focus...