The Why and How of Multi-Cloud Adoption
Hiring a Cloud Service Provider (CSP) was supposed to simplify IT. So why are enterprises doubling the hassle by hiring two… or more?
New and changing business needs, proliferating data initiatives and the good old learning curve are prompting more and more enterprises to embrace two or even all three of the major CSPs – AWS, Azure and Google Cloud. Their reasons for doing so underscore the advantages of continuously doing your homework rather than taking a “set it and forget it” approach, as well as a desire to avoid lock-in and stay agile. Common benefits of a multi-cloud strategy include the following. (This article is geared to traditional Global 2000 organizations, with a long on-premises history, rather than newer ventures that are “born in the cloud.”)
- Improved ability to meet SLAs. Depending on your requirements, there can be meaningful variation in KPIs, such as latency by CSP and geography. The ThousandEyes research firm released figures last November showing, for example, that CSPs that route traffic through the Internet more than their own backbone in regions such as Asia, can have markedly lower and more variable performance. Enterprises can use research like this to guide their choice of CSP for different business units, according to the user and hosting locations for each.
- Reduced cost. Different initiatives at the same company might find their best price at different CSPs. As with SLAs, mileage will vary by workload requirements and CSP, so there is no simple formula for cloud pricing. As an example: TechRepublic recently found the total monthly cost of 1 VM (including two virtual CPUs and 8GB of RAM), 1 TB of storage and an Active Directory Authentication service ranged from $63 to $98 across AWS, Azure and Google Cloud, with considerable variation by component. The implication is that one CSP might have a more compelling price for a high-storage, low-compute use case, while another CSP will offer the best price for low-storage, high-compute use cases. And taking the time to diversify in this way can greatly reduce costs. We should expect prices to keep evolving as the three leaders jockey for position in this fast-growing but competitive market.
- Reduced lock-in risk. Enterprises rightly fear the prospect of perpetually committing to one cloud environment. They are well advised to stay agile and ready to adjust as needed. This is best achieved with more standard applications and data sets as they can more easily be moved between clouds or back on premises as needed.
- Hedged competitive risk. As Amazon broadens its services and market presence, cloud clients in retail, financial services, healthcare and other segments are scaling back their sensitive-data exposure to cloud pioneer and leader AWS. For example, they might move their real-time analytics workloads to Azure or Google, but maintain archival and TestDev environments on AWS. This is one contributor to the recent growth of Azure and Google Cloud, which overlap competitively with fewer industries.
- Specialized support of advanced analytics. All three CSPs have invested heavily in the latest predictive technologies, as attested by AWS SageMaker, Azure Machine Learning and Google Cloud AI. Altexsoft has published a thorough comparison of these “ML as a Service” offerings, differentiating them by custom predictive analytics capabilities as well as API support for speech/text processing, image analysis and video analysis. As with each benefit above, there can be striking differences in the ability of each CSP to meet individual granular use cases and requirements. One ML initiative might benefit from Azure’s anomaly detection, recommendation and ranking algorithms, which another initiative at the same company needs Google’s extensive APIs for image analysis.
These and other benefits can outweigh the downside of working with more than one provider – namely, management overhead, administrative complexity and switching costs (for example, to migrate from one CSP to another). Figure 1 summarizes the pros and cons of a multi-cloud strategy for CIOs and their teams of IT systems administrators – typically the cloud decision makers – to consider.
Figure 1: Multi-Cloud Weighed in the Balance
Provided the upside is higher for your organization, there are several multi-CSP selection criteria to consider. These include SLA performance, pricing, competitive overlap and specialized analytics capabilities as outlined above. Additional selection criteria include the following.
- On-premises affinity. A primary question will be: how similar to my data center will this CSP environment be, and how easy will it be to migrate, resume operations and expand on their cloud? AWS has sought to provide a simple answer, for example with its RDS service for setting up and scaling familiar on-prem relational databases such as SQL Server and Oracle on their cloud. Azure has a “homefield advantage” with SQL Server, providing easy two-way bridges between on premises and Azure cloud environments and familiar SQL capabilities on either side.
- Lock-in risk. By using public or standard APIs wherever possible, you can keep your applications and data relatively mobile. Lock-in is a greater risk for more specialized services – ML as a Service, etc. – that involve custom coding with cloud-native toolsets and APIs. Organizations should carefully select the right CSP(s) for their custom requirements, understanding those decisions might be hard to unwind down the road.
With these various factors in mind, here are the primary multi-cloud scenarios or use cases enterprises are addressing.
- Change or rebalancing of CSPs. Most enterprises with a cloud presence – i.e., most enterprises – started with AWS. Now some are taking an AWS + 1 approach, for one or more of the reasons listed above. For example, a Fortune 100 food processor moved its on premises data lake to AWS about a year ago. After consulting a partner that competes with AWS, the new CIO of this firm opted to move analytics workloads and data primarily to Azure, which presents less competitive friction and high compatibility with its on-premises SQL Server environment. Archival, test/dev and read-only DB environments likely will remain on AWS.
- Cost arbitrage. As outlined above, it is well worth the time to carefully map data and workload requirements (storage, CPU, etc.) to available offerings from the big 3 CSPs, and periodically reassess. This approach best fits less-custom environments with low lock-in risk that can be easily ported between clouds.
- Diversification by initiative. Different BUs and different data/analytics initiatives often can best support their SLAs and/or advanced analytics use cases by using different CSPs. For example, a major financial services organization in Germany has selected all three major CSPs for its data modernization initiative: Google for machine learning, AWS for infrastructure efficiency and Azure for CRM operational analysis. As another example, a global software provider to the travel services industry has specialized cloud platforms for distinct corporate objectives. AWS will help with cost, infrastructure optimization and performance, while Azure will enable new AI and advanced analytics services.
Successfully navigating the multi-cloud world requires several best practices.
- Plan carefully. As should be clear by now, despite the desire to simplify IT decisions by moving to the cloud, you get the best results by sweating the details. Carefully defining application domains, and the cloud platform selection criteria for you, data teams can deliver significantly enhanced value to the business at a lower cost.
- Take a phased approach. The learning curve is critical, as many pros and cons of a given cloud choice only become obvious after an initial decision is made. By starting with one cloud, then moving certain applications and analytics workloads to another, then reassessing, you can make each successive step as well-informed as possible.
- Keep it simple. Sweating the details of your requirements for a given application domain does not mean you should have dozens of domains. Rather, start with just a few domains that merit different treatment, and drill down on those. You can generate significant overall results with just two or three diversification decisions and thereby reduce ongoing administrative tasks.
- Monitor and re-assess. It will be critical to monitor your multi-cloud success according to KPIs, and periodically reassess how the variables have changed: your workloads, business requirements, available resources, and perhaps the KPIs themselves. Periodic rebalancing will generate the best results.
Cloud computing has delivered on many promises: efficiency, economy, flexibility and new service support. Data teams might further amplify those benefits by making things a little more complicated and hiring another CSP.